Introducing: AskSteve


Collaboration and access to resources is one of the most important enablers when trying to learn or experiment with new technologies, concepts and applications.  In fact, this is one of my favorite things about working at Nutanix, having the ability to collaborate, discuss and strategize with people much smarter than myself!

This is just that, AskSteve is a way to give you direct access to myself and the resources I have access to so you can get the information or knowledge you need to be successful!

Here are some examples of topics:

  • Anything and Everything Nutanix (technology, design, best practices, troubleshooting, etc.)
  • Technologies and Trends (VMware, Microsoft, Citrix, Arista and the list goes on…)
  • Architecture and Design
  • Whatever!

Feel free to ask any question and I’ll do everything in my power to get you the answers you’re looking for in the comments below!

Also, don’t forget this is meant to be collaborative and help grow the community so always feel free to chime in with answers and comments on topics so we can have some great discussion. :)

Look forward to collaborating with you all!

  • toffendask

    Hi steven,

    Does nutanix provide simulators so we can test and get skilled at home? This is exciting stuff. Or do you use an actual server at home?


    • stevenpoitras

      Great question! Currently today we do run “virtual clusters” internally, however it isn’t something we’ve exposed as of yet…. Stay tuned :)

      • toffendask

        OK, thanks. I know that for instance NetApp provide a file you can download up fire up in Vmware and hey presto.. you can test, train in

        a real environment.

        If Nutanix get this going I d be keen.

        • stevenpoitras

          Definitely, something is in the works :)

          • MGS

            Wondering where you’re at with having this available? Thanks.


          • sideshowtob

            Would love to know also! Be great to be able to test and learn about Nutanix in my own vSphere lab.

          • stevenpoitras

            Unfortunately I don’t have any dates, but I do know this is being worked on… I will keep you guys in the loop when I know more!

  • Joe

    Hi Steven!
    I have NX3050, which came after testing by customers. To restore the default settings is not possible.
    How can I return the system configuration to the factory settings?

    • stevenpoitras

      Hi! There are actually a few ways to do this depending on what the desired outcome is (eg. cluster data/configuration removed, IP addressing changed, etc.) To “destroy” a cluster you can use the cluster -fy destroy command which will delete all of the cluster data and metadata. To do a full “reset” we have a tool called Phoenix which will do this for you. Feel free to reach out and I can add more details vmwaredude at nutanix dot com :)

      • Joe

        Helped by phoenix.iso

  • Damian Yenzi

    Steve, do u know of anyway to reach the console of the individual node within the block without deploying KVM? I want something similar to HP’s ilk or Dell’s DRAC product. Is it possible thru one of the network connections to the block? thanks

    • stevenpoitras

      Hi Damian, there is an IPMI interface which can be used to connect remotely to each node over the network. Hope that helps! :)

  • Simon

    Hi Steve, I would like to ask, Is the “Replication Factor” data protection in NDFS “Nutanix block” aware ? More specifically, For a 4 node NDFS cluster made from 2 NX6000 (dual node) blocks. Would “RF=2″ ensure data availability in the event the one of the blocks go down ?

    • stevenpoitras

      Hi Simon, great question, RF is block aware ensuring the data availability in the case a full block goes down. There are also two levels of block awareness, first within the metadata replication which is RF3 by default and then the data which is RF2

  • smolk

    Hello Steve

    I have a question regarding Nutanix dedupe engine. Would it be recommended to store non-dedup friendly data, such as vm swap, user and temp directories to a datastore without dedup enabled? Traditional storage arrays usually recommends this so that RAM and SSD doesn’t fill up with dedupe tables. For better explanation see link.

    Regards N

    • stevenpoitras

      Hi, for fingerprinting to be invoked its necessary to perform a write I/O larger than 64K currently (primarily targeted at full clones, P2V or V2V migrations to Nutanix). That will most eliminate things like swap, user or temp data from getting fingerprinted as Windows will normally use a 4K I/O size.

      In RE: to cache bloat and metadata, that’s where our multi-pool structure in the content cache comes into play as data that is only read once (“single touch”) will only be pulled into the single touch pool and then evicted from the cache. Here’s some more details on the cache structure:

      As for metadata, all of the fingerprint information is stored in Cassandra which is a massively scalable key-value store which runs on each node. This scales as the cluster grows so theres no need to worry about the size of metadata as the platform hosting it scales as the cluster grows.

      Let me know if you have any other questions!

      • smolk

        Thanks for the explanation, very well explained and a pretty clever way of handling it.

      • banc

        Hi Steve

        I understand that I/Os smaller than 64k are not
        finger-printed inline. But isn’t there a post-dedupe (fingerprinting)
        happening for all other data?


  • mr.rosh

    Hi Steve,
    it would be nice to have pdf versions of Nutanix Bible, so it’s easier for reference offline.
    Any chance for that in future updates, releases?

    • stevenpoitras


      Definitely a request I’ve been getting a lot lately!

      If you’re using the Chrome web browser you can just print the page as a .pdf and that should do the trick! Other browsers should also have this capability as long as acrobat is installed.

      I’ll also look as having a link which will export the html to a pdf as well


  • banc

    Hi Steve

    Can you give some more details around optimal block sizes for guest VM I/Os? In the Nutanix Bible you wrote that there are sub-extent read/write I/Os happening in the NDFS back-end. What size can those I/Os be? Or asked differently: If I can choose the I/O size of my application/OS what would be a good choice?


    • stevenpoitras

      Hi Sam —

      For the optimal block sizes this will really depend on the workload and what the ideal benefit is (eg. IOPS v. Throughput). For most workloads the default block size doesn’t need to be modified, for example with VDI the default 4K block size works great.

      However, for workloads which can be throughput focused (eg. SQL Server), a larger block size can help increase this so a 8K or 64K block size is ideal.

      I would also note that with us and how we handle extent and extent groups there is no need to perform alignment within the guest OS.

      Let me know if you have any other questions!

  • Michael
    • stevenpoitras

      Thanks for the feedback Michael!

      Looks like the links are broken… could you let me know what in particular you’d like to see added? (eg. diff network config options, wiring, switch configs, etc.)

  • RandomGuy

    I’ve seen this in many places (even similar info here in your Nutanix Bible under extent stores):

    “Nutanix is built to function independently of storage media.
    Besides storing data using local high-capacity hard disk drives and
    advanced solid-state flash, the Nutanix architecture is extensible. It
    will enable easy consumption of public clouds via standards-based APIs
    to build seamless hybrid cloud environments.”

    Can you elaborate? How would one go about doing this? Or is this fancy marketing language that you can use ESXi and vSphere to mount other storage as well.

    • stevenpoitras

      Good question, so essentially today we have a SSD and HDD tier for data storage. However, there isn’t anything keeping us from adding additional tiers such as an existing NAS or cloud (S3/Azure) – in fact it’s already in the code.

      For example, I some of these running, keep your ears open for more here :)

  • haroon

    Hi Steven,
    can you please tell me how many node failure can sustain cluster up and running?

    lets say i have 4, 8 and 12 nodes cluster so how many minimum nodes should be up running?

    • stevenpoitras

      Hi Haroon —

      Apologies for the slow response!

      Good question, this can be broken down into two pieces: max number of node failures, max number of concurrent node failures.

      In the back-end it all comes down to metadata and data replication factor (RF). For example, when using RF2 (N+1) for data, metadata is at RF3. When using RF3 (N+2) for data, metadata is at RF5.

      This means at RF3, I can support 2 simultaneous node failures for smaller deployments, or 2 simultaneous full block failures for 3+ block deployments.

      When a node/block/disk fails essentially the data stored on that object is unavailable and this kicks off a Curator scan which will re-replicate the now under-replicated data to maintain RF. All nodes participate in the re-replication so as the cluster grows the efficiency scales as well.

      Now, to the question of max supported failures, essentially you can have every node fail down to the point where there is still a quorum. So for a cluster using RF2 data and RF3 metadata this means you could have every node sequentially fail down to two nodes (assuming you had enough storage/compute capacity).

      For example, if you had a 24 node cluster and nodes failed sequentially (allowing data to be rebuilt), you could go all the way down to only 2 running nodes.

      For a RF3 data / RF5 metadata cluster you’d need 3 nodes to have quorum so similar to the above just going down to a min of 3 nodes.

      Hope that helps and let me know if you have any other questions!

  • hwdoc

    Hi Steven,

    I have two questions regarding how it works under the hood:

    – Is it possible for standalone ESXi servers to use the N(D)FS share of a Nutanix cluster? I heard it would, but it will only see it presented by one CVM, so it wouldn’t be redundant.

    – Can you build a Nutanix cluster with one big datastore while the hosts build two separate vSphere clusters with isolated VM-Networks (except the CVMs)?


    • stevenpoitras

      Hi Doc —

      Good questions!

      In RE: to external ESXi hosts accessing NDFS, this is possible to do today tho not necessarily recommended as part of the core value is the localization of I/Os provided by the converged nature. However, if looking to expose it’s recommended to use the external cluster IP (introduced in 3.5+ I believe) and modify the whitelist to allow traffic from the external ESXi host’s IPs

      In RE: to a large Nutanix cluster with a single datastore exposed to multiple vSphere clusters, you can absolutely do that (I actually do this in my lab)! As long as the VMkernel and external CVM interfaces are on the same network, the vSphere clusters can have different VM networks

      Let me know if that helps or if you have any other questions!

      • hwdoc

        Hi Steve,

        thank you for your reply, this helps a lot!

        So using N(D)FS from traditional ESXi hosts (maybe in a separate vSphere cluster) via the cluster’s storage IP would mean to have a
        redundant path to the storage for those hosts?
        The data would just be distributed like the one of the Nutanix hosts according to the replication factor that has been set for the Nutanix cluster but without using the (performance) advantages of having local data on the external host. Am I right?

        Just to let you know: We want to upgrade our vSphere environment with Nutanix systems which hast two clusters on every site: One for internal systems and one for DMZ, all attached to a traditional SAN atm which also has to be replaced.
        The DMZ cluster consists of only two ESXi servers and the performance requirements aren’t too high so it would be great to use those hosts for one or two more years and simply connect it to the Nutanix storage built with new hardware where we need performance (internal cluster).

        Would you say this will be a supported solution?


        • hwdoc

          Hi Steve,

          could you please take a look at my last questions?


          • stevenpoitras

            Hi Doc —

            It would definitely work, however I can’t comment on whether or not it would be a “supported” configuration, but those are grey areas which are normally pretty easy to work around :)

          • hwdoc

            Thanks for your reply! :)

  • banc

    Hi Steven

    I had a discussion recently with a customer around the necessity to always have enough storage available in case a node would fail (n+1). The customer would like to use more than 66% of the available space in a three node cluster since he is also replicating to a second cluster. So in case of a second node failure he would just failover to the secondary site.

    This discussion made me think of how exactly NOS behaves in case of a node failure if there is not enough physical space to re-protect all data. Can you help me with the following questions:

    – When the cluster starts to re-protect is there a threshold (let’s say 95% used physical capacity) where the replication (curator) stops to give priority to new blocks being written or will the cluster just fill all remaining space to re-protect existing data?

    – What happens if there is still space available (let’s say replication is running but did not yet fill up all space) and new blocks get allocated by a VM? Will the cluster allow new blocks being written even though it’s clear that a full re-protection will not be possible due to too little physical space? Or will it only allow writes to existing/previously written blocks?

    – If a VM would allocate new blocks (I am not saying the VMDK gets extended) in the already provisioned VMDK/Vdisk while there is no free physical space left. Meaning writing to blocks it has never written before then the ESX host will get a “VAAI Thin Provisioning Stun” from the cluster in case of vSphere, right? So basically the VM gets frozen. Do you know what happens in case of Hyper-V and KVM? Are there similar mechanisms?

    Your answers are much appreciated!

    • stevenpoitras

      Sam —

      Great questions!

      First and foremost, it’s always recommended to keep enough storage capacity to allow for new ingest data as well as re-protection in the case of a node failure. No one wants data in an un-protected state :)

      Here’s a simple formula to help calculate that with some buffer:
      cluster usage < 90%[cluster capacity – max node's capacity]

      cluster usage < 90%[cluster capacity – max node's capacity – second max node's capacity]

      In re: to your question on how the system reacts when re-replication and write I/O are being performed and fill the storage up, essentially it will allow both to continue until the storage capacity becomes 95%. Which, at that point will not allow any more writes or re-replication of data.

      If you were to remove some older data (eg. un-needed VMs or old snapshots) both would then continue to proceed until the 95% limit is reached.

      For larger clusters this becomes less of an issue as the required % of storage needed for n+1 goes down as the number of nodes go up (eg. 25% for 4 nodes, 12.5% for 8 nodes, 6.25% for 16 nodes, etc.)

      The moral here is to ensure that you always have enough storage capacity to allow for a single node failure and still have a buffer for new incoming writes then just scale out with new nodes to increase the aggregate size of the pool. :)

      In re: to any Hyper-V or KVM equivalent to a Stun, I'd need to do some research and find out more there!

      • banc

        Hi Steven

        Thanks for the detailed answer! Absolutely makes sense. Maybe you can also help me with the following question:

        When I open the Storage and table view in Prism to display all containers I see several columns. One of them is “FREE RESERVED”. Another one is “RESERVED CAPACITY”. Some of the containers show a value greater than 0 even though we do not use any reservations. According to the online help it is only applicable if reservations are used to my understanding.


  • hwdoc

    Hi Steve,

    within the last days two more questions crossed my mind:
    – Is there a document explaining the Integration from Nutanix into a VMWare SRM environment? As I understand, the use of SRM is an option due to have an integrated replication as well.
    And building up on this:
    – Is there a recommendation you would make concerning datastores/storage units?
    ATM we have tradional SAN-LUNs: only some of them are replicated to another site via SRA/SRM so every VM on a replicated LUN has to be licensed in SRM. Less important VM’s data won’t be replicated which should remain the prefered way to handle that.

    Would this mean to create three datastore per cluster:
    One for less important VMs, one for important VMs (replication source) and one for replicated VMs (destination) from the other cluster?

    Thanks in advance!

    • stevenpoitras

      Hey Doc —

      I do know there is a best practices guide which is currently being worked on which should cover the integration and optimal config settings. I don’t have a firm date as of yet but I’ll check on this!

      I’ll admit I don’t know a ton about SRM, however all of the replication from a Nutanix standpoint is done via Protection Domains and Consistency Groups. A PD is essentially a group of VMs to be replicated and the CG are VMs that get snapped together in a crash consistent manner.

      With this construct you can still have one single massive NDFS datastore and then for the VMs you want to replicate and protect via SRM you’d just have a PD for those. (Assuming SRM doesn’t replicate a full datastore).

      I’ll do some more digging here but hopefully that helps in the interim!

      • hwdoc

        Hey Steve,

        thanks for the answers!

        SRM in conjunction with NetApp-RAW-Storage and its SRA would mean to have a complete LUN replicated, not just based on VM.
        Do you know how Nutanix’ SRA works in case of this?

        So atm we use dedicated LUNs for replicated VMs also because SRM would report an error when finding a VM in a protected datastore which is not configured (and so: licesed) by it.
        May be cards are shuffled new when using SRM 5.5 on a datastore reported as NFS but I think it is not; must also be a question of how the SRA works.

        Hope to hearing from you soon!

  • zapbranighan

    Hi Steve,

    I know that most modern VMware configs use VAAI (ATS), but in the case that it doesn’t how does Nutanix support SCSI-2 and SCSI-3 (Persistent) Reservations? Reads and writes can come from any Nutanix Node, how are reservations managed?


    • stevenpoitras

      Hi Zap —

      For VMware environments NFS storage (w/ NAS VAAI support) is used and doesn’t require the ATS primitive for VMFS. Hope that helps!

      • stevenpoitras

        Also, to shed a little more light, the local CVM has a exclusive lock for all of the VMs/vDisks running on it (NDFS context). This eliminates any sort of lock contention. In the event that a VM is migrated to another node the now local CVM would take over the lock(s).

  • JL

    Hi Steve,

    NX3360 , is running a mixture of VMs from server 2003 ,2008 ,2012, and a total of 270 vms across the cluster, The Vms is performing (simulated) as real work enviroment, the problem is the Disk latency on both Read and write, Is very high not what i am expecting. I did another test only run 80 Vms on 1 node to see only Read I/O to the Disk and the AVERAGE latency on read was 11ms and the problem is the intermittent 100ms now and then , there is Database more than one running, any suggestions, Both the PY and Iperf test is showing good results as expected.

    what could cause disk latency on Read (SSD is 33% only)



    • stevenpoitras

      Hey JL —

      There’s a lot of variables to track here :)

      If you can shoot me an email with the following I’ll take a look! My email is (perks of being an early employee :P)

      – Workload type
      – Avg. I/O size
      – Avg. Queue depth
      – I/O type (random/sequential)
      – Features enabled on container (eg. compression, dedup, etc.)
      – Network connectivity (1Gb/10Gb)
      – Provisioned via VAAI?
      – Storage used
      – Using VMware snapshots?
      – CPU usage?
      – Memory usage?
      – VM specs (vCPU/MEM/Disk)
      – NOS version

      With 270 for a 3 node clusters that’s ~90 VMs per node on average. Using normal server sizing 2-6 vCPUs, 2-4 GB mem, and 40-80GB disk, I’d best most of the compute resources are fairly constrained.

  • Matt Davies

    Hi Steve,
    I have a 6 node cluster running VMware that we are in the process of migrating to Hyper-V. We have removed three nodes from the Nutanix cluster and VMware cluster and are in the process of using the Foundation tool to reimage them with 4.0.1 and HyperV however the tool never detects any nodes to image. Any ideas?

    • stevenpoitras

      Hey Matt —

      Were the nodes remove using the dynamic remove node tool? Besides via Prism another way to validate the nodes are fully removed from the cluster is to look at the tail of the ~/data/logs/genesis.out file and ensure it says “Waiting to be configured via RPC”. If it doesn’t you can always run ‘cluster -s localhost -fy destroy’ on the node to ensure they’re ready to be re-imaged.

      Also, from a networking config does the laptop or VM running Foundation have direct L2 connectivity to all of the hosts on the IPMI network? Essentially the hosts will PXE boot from the Foundation VM so a direct connection to the network will be necessary (depending on Virtual Box or virt software running the VM) as some use NAT by default which won’t work.

      Also, in most cases it’s ideal to have the nodes and laptop on the same switch, as some networks with STP won’t converge fast enough (RSTP can help here).

      Let me know how it goes and feel free to reach out with more questions!

      • Matt Davies

        Hi Steve,
        Thanks for this.
        The nodes were removed correctly (a Nutanix support person did it), I need to check the genesis log. There was a L2 connectivity issue which I have resolved and now I see the 6000 in foundations however the 2x 3050’s don’t show up.

        • stevenpoitras

          Hi Matt —

          Did you ever get this resolved? If you’re unable to see the 6000’s I’d guess the cables might be hooked up to the wrong port.

        • Matt Day

          Are the 2 new 3050 Nodes on the same subnet as the Foundation VM? Also as Steve mentioned you need to have them

      • Matt Davies

        I have just added the two 3050’s in as bare metal however when I go to image them along with the detected 6000 the installer complains that the hypervisor image is not supported, even thought it is the correct 2012 R2 VL version.

  • Guest

    Hi Steve,
    I have a 6 node cluster running VMware that we are in the process of migrating to Hyper-V. I have removed 3 nodes from the Nutanix and VMware cluster, I have deployed the foundations VM and was about to re-image but I don’t see any available nodes in the tool. Any idea?

  • zapbranighan

    Hi Steve,
    Is there a reason why Nutanix doesn’t use erasure coding instead of, or in addition to, replication factors? MIrroring can eat up a lot of storage based on the size of the volume and/or high RF numbers. I understand that mirroring is much simpler, but one can argue that erasure coding (FEC) is a good middle ground between saving storage and complexity for web scale storage.


    • stevenpoitras

      Hey Zap —

      Absolutely great points, as known all of our features are delivered via software so software updates can include new features like erasure coding. I’ll leave my non-response as a hint :)

  • hwdoc

    Hey Steve,

    one question – regarding networking and VMWare vDS:

    We definitely need one physical 10 GbE-Port from each node for
    mirroring VMs’s traffic to an external monitoring system. So what’s left are
    one 10 GbE-Port and two GBit-Ports per node and two core switches in our case.

    Is it possible to build a redundant network setup for each node with this and how would you suggest to build this with VMWare’s vDS?


    • hwdoc

      Hi Steve,

      could you please take a look an my question regarding vDS?

    • stevenpoitras

      Hey Doc —

      What I’d do is configure the vDS with the 10GbE as the primary active uplink and the 2 x 1GbE uplinks as passive / standby adapters. This will keep the primary traffic on the 10GbE but will still provide redundancy if the link / switch were to fail.

      Also, if possible, I’d still configure the mirrored 10GbE as a standby adapter if possible to eliminate any potential perf degradations if the primary 10GbE link were to go down.

      Hope that helps!

      • hwdoc

        Steve, many thanks for your advice!

        Can you tell from your experience: if we limited all traffic to just one 10 GbE-link – will we generate a _noticable_ performance bottleneck regarding all kinds of network flows like

        – NTX storage
        – VM-traffic
        – NTX replication to DR site

        Our setup would be about eight NX-3060 on one site and four on a second site, 150/75 VMs with all kinds of workloads (SBC, application-, SQL-, Webservers).
        So what bandwidth do you see looking at other customers and did you ever see a setup runnig into problems having limited bandwidth?

        BTW: NX-1020 is limited to 2* (max: 4* GBit) – how does this scale?

        What about the upcoming NX-1030?


        • stevenpoitras

          Hi Doc —

          Sorry for the late response, busy times with VMworld :)

          It all comes down to the amount of write I/O and the I/O between the apps / VMs, but in most cases I’ve seen it very difficult to fully saturate a 10Gb link

          Here’s a quick estimator I put together to help calculate:

          For the 1020 it’s primarily targeted towards smaller branch office where the I/O load is fairly light so the 1Gb connections should be able to handle the network BW

          Let me know if you have any other questions!

  • Alexis Lorca

    Hi Steve, I have a 3 ESXi and 3 KVM nodes cluster 4.0. I have a VMware HA and don’t want to license the other 3 nodes. I split this cluster on 2… And today can’t create an iscsi container. Don’t know why but on the KVM cluster can only create NFS containers.
    I delete all containers and create the storage pool from scratch but can’t create the container compatible with KVM.
    Any hint?

    • stevenpoitras

      Hi Alexis —

      Are the 6 nodes in a single Nutanix cluster? If so all you need is the single NFS datastore and the 3 KVM can server storage for the nodes running ESXi.

      If the clusters are separate (what it sounds like) KVM uses direct attached iSCSI disks to the VMs (no need to create a iSCSI datastore). If you take a look at the 3.5 Platform Administration Guide it provides some examples on how to use the commands to provision VMs, etc.

      Let me know if you have any other questions!

      • Alexis Lorca

        Indeed, I have 2 clusters today. I remove the KVM nodes from the cluster and those nodes form the second cluster.
        But can’t present the container to the KVM, because it says that there are no NFS compatible nodes on the cluster.
        So I think that there should be a kind of mark on the nodes that disable the iscsi possibility.
        I don’t want to reflash the nodes, I’m sure it could work, but I prefer to have the chance to learn from this situation.
        Does it make sense to you? Have to see something similar before?

        • stevenpoitras

          Hi Alexis —

          Sorry for the late reply!

          The UI will say that there are no compatible NFS nodes on the cluster, however, all containers are exported as NFS and SMB by default. The one thing you’d need to do is add a whitelist entry for the IP range trying to access the container

          So you can do the following:

          1. Create the container (if one already doesn’t exist)
          2. Create a whitelist for the IP range wanting to access the container
          3. Browse the container on the remote cluster by using SCP / Explorer / NFS. The IP will be an IP of any CVM or the cluster IP (preferred) and the share will be the container name. Example: \

          Let me know if you have any other questions!

  • Bill Lynn

    Steve, I am looking at the possibility of running Nutanix on a shared infrastructure system from Dell. The problem is it shares a RAID controller between 4 blades. The blades have their own local boot devices but the data store is a shared RAID controller. Is there a way I can configure the storage to run with LDs instead of PDs?

    • stevenpoitras

      Hi Bill —

      From a product standpoint we don’t rely on any RAID constructs/controllers as the data placement and protection is handled within the software (also, don’t want to have a single point of failure in a RAID controller).

      With that said the CVM’s OS will detect disks which are exposed to it so it would be theoretically possible. SSDs and HDDs are required to be as separate devices allowing the software to control the logic. So if you had LDs for the SSD devices and HDD devices the OS would see them as a normal device (some modification would likely be required for this to work correctly)

      Feel free to shoot me an email and we can discuss further! thedude at nutanix dot com

  • Jeremy Jensen

    Steve, my company owns 6 Nutanix Blocks (NX2000’s & 3000’s mixed) and have recently started using them for very large SQL Servers. Once such VM is at 5.7TB of provisioned space which actually exceeds the local storage available in our NX3000 configurations. How does the Nutanix software balance data locality when a particular VM exceeds local storage on a node?

    • stevenpoitras

      Hi Jeremy —

      Great questions, all of the tiers are pooled between the nodes (eg. Cluster SSD tier, Cluster HDD tier) with logic for which data is local. Basically it can be summarized as all new write I/O will always hit the local SSD devices. For colder/older data that might sit on a remote nodes SSD/HDD devices which will be accessed over the network. In the case that data is heavily accessed the data locality will pull it local so any subsequent access will occur locally

      Here are a few resources which explain this in some detail:

      Let me know if you have any other questions!

  • Steve

    I have a unique situation where my customer tried to upgrade their box and has seemingly bricked it. It hung during the upgrade and now won’t start. I was able to get all of the nodes upgraded. But now, when I try to start the cluster all I get is the error that genesis isn’t running on the nodes even though when I check the status the genesis service is running on all nodes. Is there any way to manually delete the cluster configuration and then rebuild the cluster? We have support, so my next step is to get them engaged. This is pre-production so I have been curious into fixing myself and only contacting support as a last resort.

    • stevenpoitras

      Hey Steve —

      Interesting, do you know when version it was running and then which version it was upgrading to?

      I’d first try to restart genesis on all of the nodes which can be done using the following command run from any CVM:

      for i in `svmips`;do ssh $i source /etc/profile > /dev/null 2>&1; genesis restart;done

      From there try the cluster start command. Also, what is the output in the genesis.out file located in ~/data/logs/ ?

      Support is here to help so I’d recommend getting them engaged as well!

      Hope that helps

  • banc

    Hi Steve

    I was reading your new entry in the Nutanix bible regarding disk breakdown and have a question about the formula to calculate the OpLog space:
    The formula for a NX3050 in Excel would be MIN(100/2;30%/2* remainingSpace ). Now I am not sure what the variable “remaingSpace” is? If it is the remaining space of the SSD (Extent Store) then I wouldn’t know the value as it depends on the OpLog size. So I changed the formula to MIN(100/2;30%/2*(2*373)). The (2*373) represents the total SSD storage. I am not sure if that would be correct though. Am I correctly interpreting the sum of all extent store space (SSD and SATA) would be the useable cluster capacity?

    Thanks for your help!

    • stevenpoitras

      Sam —

      Good question, the size is evaluated from the top down, eg. Nutanix Home -> Cassandra -> OpLog -> Content Cache -> Extent Store

      So for example the remaining space for the oplog calculation would be the SSD size in GiB – Nutanix Home – Cassandra (if SSD1) . Then the remaining GiB for extent store would be that number – Content cache

      I’ll add some details to make this clearer!

  • Steve57

    Hi Steven,

    I read your posts and the Nutanix bible and they were really helpful to understand the Nutanix Architecture. I have still some doubts about the way caching and deduplication are managed.

    In particular my concern is related to the fault of a Nutanix node. In this situation a relevant amount of VM must be restarted on the remaining nodes by VMware HA. In a standard3 tier environment this event implies an awful lot of IO from the external storage to reboot each VM. We do not have the help of the linked clones as it happens with VMware View so the time to wait for completion could potentially be very long.

    I am trying to guess if the Nutanix caching and elastic dedupe technology might help.
    Supposing for example that all the VM are Red HAT they share a large part of their images. So the question is what will it happen in term of cache usage and disk I/O when you restart all the VM that were running on the failed node?.

    This is an objection that could be raised from the supporters of traditional storage with a lot of disk drives where you can distribure the load….

    Thanks for your help,


    • stevenpoitras

      Stefano —

      First, great to hear it’s helpful! If you have any suggestions / areas you’d like to see let me know!

      Correct, during a node failure / HA event the nodes will be restarted on other nodes throughout the cluster.

      It all comes down to the type of VM and how it was cloned / created:

      Traditional Linked clones -> With shadow clones the replica will already be cached on the target node (assuming other linked clones are running there), this will allow all read I/O of the base image to be read from the local memory cache.

      Traditional VAAI clones -> With VAAI clones the common / base image used for the cloning can be cahced in each CVM’s extent cache. Assuming there are already VMs running on the target node where the VM is restarted on the majority of the read I/O for the base image can be served from the local memory cache. In the case where no VMs are running on the target node where the VM is restarted on the initial reads will be forwarded to a remote nodes memory cache or extent store (for data not in the cache)

      V2V or VMs using dedup -> When using V2V it’s recommended to turn on fingerprinting aka dedup. With any VMs using dedup each fingerprint will be stored in the single touch pool of content cache upon first read and will subsequently be moved into the multi touch pool of the content cache upon any subsequent read which spans both memory and SSD. Assuming there are VMs running on the node where the VM is restarted on it is highly likely that the data for the OS image will already be sitting in the content cache and any read requests for the OS will come directly from local memory or SSD. Similar to the example for VAAI clones if there are no VMs running where the VM is restarted the initial reads will be forwarded to other CVM’s content caches and will be pulled local.

      So from the above, really no matter what VM / clone type it is, it is very likely that all of the read I/O for boot will be hitting the local memory or SSD cache and not traverse the network. In the worst case scenario where it isn’t in the local cache it will most likely hit the other CVM’s caches and then be cached locally.

      In re: to the comment of distribution, the key reason for requiring a lot of spindle count is that traditional spinning drives can’t facilitate a lot of IOPS, however, this changes significantly with SSDs. As a single SSD can provide more IOPS than 10’s of spinning disks

      As always, all write I/O will occur on the local SSD when it is started.

      Hope that helps and let me know if you have any other questions!

      • Steve57


        the question was specific to the behaviour in a traditional VMware server virtualization environment. Without dedupe in case we have for example 40-50 VM all running Red Hat Linux, each VM ,even if very similar to the others, will be stored in a specific area of storage. At boot time VMware will have to read each VM from the HD and it will take a long time. With Dedupe on each unique segment will be stored once on disk and the metadata will associate it to each different VM.
        At boot time the first VM will be read from disk into cache but what will happen from the second VM on? As I understand the CVM can not check the fingerprints since each segment has not be read yet. Instead it should use the address of the segment as a key to check in the metadata if that specific segment has already been cached. Is this that happens or I am missing something?

        Thanks in advance for your help,


        • stevenpoitras

          Hi Stefano —

          Even without dedup if the VMs are cloned from a base template (eg. deploy from template or a normal vSphere cloning operation) VAAI will be used.

          When VAAI is used the base vDisk is marked read only and a new delta vDisk is created for each cloned VM (all of which inherit the previous block map of the base image via the copy block map operation). More here on how this process works:

          As data is read it will be pulled into the extent cache and as the VMs share the block map any VMs reading the same blocks can read them directly from the cache.

          So to put it shortly, even when dedup isn’t used any VM / Clone created natively will share the block map and hence share the same blocks in the cache

  • Fireball

    Hi Steve,

    I have the famous hen and egg question for you. :-)

    I’ve read through your Nutanix Bible and understood that the CVM, which presents NDFS to the hypervisor, itself also runs on the hypervisor. Now, on what storage does the CVM run? It cannot be installed on NDFS until the CVM runs, basically, am I right? So do the Nutanix nodes need to have local disks (such as a simple RAID-1), which are not usable for NDFS, in order to get the CVM going first?

    Thanks and regards,

    • stevenpoitras

      Hi Jonas —

      All of the real data for the CVM (aka Nutanix Home) is protected using mdadm raid 1 between a partition on each of the local SSDs which is directly mounted in the CVM. The SATA DOM contains the CVM files itself (.vmx, etc.) which are just really a shell and don’t contain the actual OS

  • Rafael Lara

    Hi steve,

    It is possible to create a Windows File Server Cluster (2012) using nutanix with VMWare ESXi or Hyper-V?. I saw your blog for creating a failover with a file share witness, but i don’t know if its possible to create a File server cluster for user data Storage (windows roaming profile and users data (folder redirection))

    We want to implement a new project(with ONLY nutanix) and one of the main goals is to migrate the file server (right now is a netapp FAS using the CIFS module).

    So, is this possible? do you have any suggestions to do this?


    Rafael Lara

    • stevenpoitras

      Hi Rafael —

      Good question. It will all come down to RPO and the interfaces for this.

      There are really two main options: MS DFS and some flavour of ZFS (eg. FreeNAS) running on Nutanix.

      With DFS you can rely on DFS-r to handle the data replication between the DFS servers and DFS-N to ensure the namespace is uniform. On that note we actually just released a DFS best practices guide:

      With using traditional DFS-r that will be using asynchronous replication which would provide an RPO > 0. If you need one = 0 then thats where a cluster would be required.

      The other option is to deploy ZFS / FreeNAS on our platform. This is one option that I’ve been digging into recently as from a filesystem perspective it does have some benefits / advantages over DFS. But it is Linux based so might not be an option if you’re a Windows shop. I’m hoping to have something out here in the next few months.

      Eventually we’ll work on getting something natively integrated into the platform so no add’l work to deploy the solution would be necessary.

      Hope that helps and let me know if you have any follow up questions!

      • Eva

        Hi Steven,

        I just wanted to follow up on this file server cluster on nutanix, has there been any developing or testing yet to incoporate this in the nutanix platform? We are seeking for something similar like Rafael.

        • scottath

          Hey Steven/Eva,

          Has there been any further progress on this topic?
          We have just received our Nutanix servers and were looking at consolidating some file servers rather than doing a P2V on them in addition to the other working servers.
          At the same time looking to get some HA/transparent failover going too.
          Mixed Windows/OSX environment currently accessing S2008R2 servers.


          • stevenpoitras

            Hey guys —

            I know we “announced” the vNAS product at .Next and I believe it’s targeted for 4.5.x (EOY)

            I’ll keep checking around!

  • Racha

    Hi Steve,

    Is it possible to run the “ncc health_checks run_all” command from the Nutanix powershell cmdlets? Or do you know of any bash scripts/powershell scripts that can run the ncc health checks and then email the results.


    • stevenpoitras

      Hi Racha —

      It is possible, but not using the native cmdlets today.

      I worked around this using the SSH-Sessions snappin from

      The good thing is that NCC will become more integrated with Prism and you can run and email on a schedule all via Prism. As part of this the NCC health checks will also be available via the REST api and cmdlets

      Here’s an very messy example running NCC and emailing the results:



      ## Script: Run and email NCC results

      ## Author: Steven Poitras

      ## Description: Run and email NCC results

      ## Language: PowerShell



      # Import SSH module to connect to Nutanix via SSH

      # Source:

      # Make sure it is placed in you path, which can be found using: $env:PSModulePath -split ‘;’

      Import-Module SSH-Sessions

      # SSH inputs

      $sshUser = ‘user’

      $nutanixServer = “”

      $keyFile = ‘Path to key file’

      # Connect to Nutanix cluster via SSH

      New-SshSession -ComputerName $nutanixServer -Username $sshUser -KeyFile $keyFile

      Write-Host “Running ncc commands…”

      Invoke-SshCommand -InvokeOnAll -Command “source /etc/profile > /dev/null 2>&1; `

      ncc health_checks run_all;”

      Write-Host “Getting ncc log output…”

      $nccOutput = Invoke-SshCommand -InvokeOnAll -Command “cat /home/nutanix/data/logs/ncc-output-latest.log”

      Write-Host “Sending email with NCC results…”

      $EmFrom = “”

      $username = “”

      $EmTo = “”

      $smtpServer = “”

      $port = 587

      $Subj = “NCC Results – $(Get-Date -Format MM-dd-hh:mm)”

      $Bod = “$($nccOutput)”

      #$Att = “c:Filename.FileType”

      $cred = New-Object -TypeName System.Management.Automation.PSCredential -ArgumentList $username, $(Read-Host “Please enter password: ” -AsSecureString)

      Send-MailMessage -To $EmTo -From $EmFrom -Body $Bod -Subject $Subj -SmtpServer $smtpServer -port $port -UseSsl -Credential $cred

  • Manuel

    Hi Steve

    I have a quick question concerning MSSQL Server with Nutanix.
    We have two sites for a HA-Cluster running an Microsoft SQL Server. The SQL Server should be replicated along those sites. A client which is connected to the SQL Server should not be disconnected (a short freeze would be acceptable as Long as the application it self does not crash) if the primary site goes down. Is this achieveable via asynchronous replication and if so do you have any information about it? And how will this all integrate with Microsofts AlwaysOn Availability Groups?

    With Metro Availability coming in NOS 4.1: How will this all change if Metro Availability is used?
    Thanks in advance!

    • stevenpoitras

      Hi Manual —

      Good question. For this I’d leverage AlwaysOn Availability Groups with a file share based witness. More info here:

      With AlwaysOn the data will be replicated between the SQL servers (a high-speed WAN link is required to do this with good performance for pure synchronous replication of the AG). Otherwise you can do it asynchronously if you can tolerate an RPO > 0.

      With Metro Availability you’d have a vSphere cluster which spanned the two datacenters / site locations. In the case of a site failure, the SQL server from Site A would be restarted on Site B via an HA event. With this you can actually have a single SQL server and rely on vSphere HA for the recovery.

      In the end it all comes down to the exact RTO / RPO req’s and where the protection is handled. With AlwaysOn the application is ensuring protection, whereas with Metro Availability vSphere HA and Nutanix stretch cluster are handling it.

      Feel free to reach out with any other questions!

  • Gabe

    Hey Steve, just learned about your site. Does your bible come in a PDF that I could print out and put in a binder? I can print out the web version of the Bible but it doesn’t format into Chapter, Table of Contents and pages very well.

    • stevenpoitras

      Hey Gabe —

      Great point and one that I’ve been hearing a lot. What I might try to do is change the format into a static page where I can do some nicer formatting. Hopefully I should have something for you here shortly!

      • stevenpoitras

        Hey guys —

        To give you an update here I’ve been working on a publishing platform that will allow me to publish as pdf, epub, mobi, etc. so you can have a printable pdf or ebook for kindle and ibook

        More to come here! :)

  • pradeepjoy

    sir, how we can calculate the usable hardisk size for Nutanix ? is there a sizing calculator available?

  • Felix


    How many write replicas(depend on RF) does nutanix support and recommendations?

    How does nutanix write performance when more RF defined in large cluster?I think the write performance/latency maybe a little slowing down in theroy.

    Because more RF means more write replica which would be replicated to another block/node in Cassandra ring through backend 10GbE switch.

    Hope get your reply,thanks a lot.

    • stevenpoitras

      Hi Felix —

      Nutanix currently supports RF2 and RF3 for data (which corresponds to RF3 and RF5 for metadata respectively).

      The decision on RF will really depend on how many concurrent failures you want to tolerate (e.g. N+1, N+2, etc.) and the workload you’re running.

      I’ve actually done testing in my environment and the performance hit is negligible for writes and in most cases not noticeable to the app. Read performance will operate the exact same in both cases

      Hope that helps and let me know if you have any other questions!

      • Felix

        Got it,thank you!

  • banc

    Hi Steve

    With vSphere 6 “getting closer” I was wondering what Nutanix plans are for Vvol integration? I get that a lot of the functionality and benefits are already there with NOS today and it’s not as big a deal compared to traditional storage solutions but I was wondering if this existing NOS technology will get the Vvols API to be able to manage NOS features from within vSphere (e.g. snapshots, replication, dedupe etc.)?

    Thanks for your reply!

    • stevenpoitras

      Hi Sam —

      Definitely something we’re taking a look at and will look to integrate with. The key benefit would be per volume I/Os, other than that we’re already doing all of the VM / vdisk level items.

      Will have more to come here in the future.


  • Frederic Lhoest

    Steve, will you be @ the .Next event next June ?

    • stevenpoitras

      Hi Frederic —

      I’ll be there and giving a few sessions as well!

      Looking forward to meeting everyone!


  • banc

    Hi Steve

    With Nutanix models using more than two SSDs, would the “Nutanix Home” partition still be on all of them or only the first two? I am thinking about the information found here:


    • stevenpoitras

      Hi Sam —

      Good question, it would just be on the first two SSDs, I’ll add some updates here!

      • banc

        Thanks Steve! And what about content cache (also 20GB?) and OpLog calculation (is that the same formula)?

  • banc

    Hi Steve, this is a topic we are also very interested in. We did an OpenStack & Nutanix installation about a year ago and it wasn’t well integrated to the point where it was almost unuseable. Recently I got the information that Nutanix is working on a Cinder and Nova integration but that there is no commited road map. On your website however ( it says:

    “Nutanix provides a single scalable data fabric which integrates with all of the OpenStack services, including a Nutanix Cinder backend for snapshots and volume management, as well as unique integration with Nova which allows for faster provisioning of OpenStack guest instances.”

    Can you tell us where we can download the necessary software?


  • Nick

    Hi Steve

    With VCAI – you qualify the following:

    These scenarios apply to VMware View:

    View Full Clone (Template with Snapshot) –> VAAI will NOT be used

    View Full Clone (Template w/o Snapshot) –> VAAI WILL be used

    View Linked Clone (VCAI) –> VAAI WILL be used
    What about VCAI-VAAI when creating a second replica from a Gold Master VM with two or more snapshots (i.e. recomposing a pool based on a change to the master)? I see storage efficiencies when recomposing from the first snapshot in the GM's tree, but not the others.
    Thanks :)

  • Chris

    Hi Steve,

    In reference to your post

    Could I install the Failover Cluster role/feature on my VM template so that it’s already installed as a part of the image ? Then only configure it on those VMs that will participate in an AG?

    • stevenpoitras

      Hi Chris —

      Absolutely, you definitely could install the failover cluster role / feature in your base image in a non-configured state. I actually do this for my SQL VM templates

      Hope that helps!

  • tenKe

    Hi Steve,

    Why does Nutanix in its Best Practices always recommend to use single container for all VM? Is there any performance advantages here?

    • stevenpoitras

      Hi tenKe —

      It’s all about simplicity here and not making things complex for the sake of being complex. It’s preferred to keep a single container just for that reason.

      With that said there might be scenarios where multiple containers are necessary (e.g. multiple tenants, container for dedup / compression, etc.)


  • Roman Lopez

    Here are some regexes I found useful when extracting the fields from several of nutanix’s syslog messages through splunk:

    ^(?:[^ n]* ){2}(?P[^:]+):s
    ^(?:[^ n]* ){3}[A-Z]+ ([d^:]*),d+ (?P.+)$
    ^(?:[^ n]* ){3}(?P[A-Z0-9]+)s

    • stevenpoitras

      Awesome! I’m actually working on a Splunk on Nutanix KVM RA so I’ll be sure to get these in there!

  • Dieter

    Hi Steve,
    how can I find out whether or not the layout of our Nutanix system is suiting our needs?
    We have really hard performance issues here.
    Kind regards,

    • stevenpoitras

      Hi Dieter —

      Have you contacted Nutanix support and your local SE? They should be able to resolve any issues you might have.

      If any problems persist, let me know and we can figure it out!

  • George Neville

    Hi Steve,

    We have a customer request and below is the list of requirement and we would like to know whether its possible to perform it in Nutanix nodes

    Possibility of having additional 2 1G Ports on NX-8035-G4 / NX 6035C Series

    Possibility to order Quad 10G Base T instead of Quad 10G SFP+ on NX-8035-G4

    Many Thanks

  • Jesse L. Ferguson II

    Steve, I had issues during an install and when it completed I realized KVM was used instead of ESXi. Something went wrong with Foundation. I can’t find direction on how to proceed with reinstalling using the proper hypervisor. Do you have any advice?

  • Pawel Buszczyk

    Hello Steve,

    Firstly I would like to thank you for the Nutanix Bible, it was really helpful to understand the Nutanix Architecture.

    I only have a question about rack-awareness. I can’t find any information which confirm rack-awareness is implemented. Please advice.

    Pawel Buszczyk

    • stevenpoitras

      Hi Pawel —

      Today we only do “block awareness” within the product, however you can get “rack awareness” by striping the blocks in the cluster across racks.

      For this deployment I’d recommend using at least 3 racks and FT2 so that the replicas can be distributed between them. Then I’d just make sure the zookeeper nodes are 2 in rack 1, 2 in rack 2, and 1 in rack 3 (basically don’t want 3 in one rack)

      Another option is to have a cluster between racks and run metro between them

      Let me know if that helps or if you have any other questions!


  • Bluecrest

    Hi Steve,

    Maybe a stupid question but I need to ask, as here in South Africa we have what is called “Load Shedding” by our electricity supplier. This means that the country is on a rotational power availability grid during heavy usage. Sometime the outage can be up to 4 hours or longer.

    On a smaller site where a smart UPS would take over during grid failure but the power grid not come up in time to alleviate the power problem, is Nutanix able to receive information from a smart UPS system (in good time) and replicate and hand over all data to another site as well as power down the local Nodes / Clusters until the power grid is back up for a pre-determined time?

    Hope that makes sense?

    …Steve Nagel

    • stevenpoitras

      Hi Steve —

      Really good question, and very interesting use case.

      There isn’t anything built in to the product to do this orchestration which is exposed, however every action within Nutanix is programmable (replication, failover, etc.)

      One way you could achieve this would be to have the power outage / alert trigger a workflow which would talk to the Nutanix APIs to automate the replication and site cutover.

      If you need any help with the API calls necessary let me know and we can get a solution working!


  • Jeremy Mayfield

    VMware VUM Patching and Nutanix: A lot of documentation is out there concerning the one click update process but there’s really nothing out there about VUM updates. Has anyone covered the topic or written a script that is able to patch using VUM that shuts down the cvm, patches 1 node at time then brings up the CVM before moving on to the next node?

    • stevenpoitras

      The issue with VUM is that it will fail unless you power-down the CVM as it can’t be migrated to another host.

      A much easier solution is updating the hosts using offline bundles (if you can’t use the upgrade built-in to Prism or have other packages that need to be updated)

      Using the Nutanix ‘hostips’ variable you can have the command update all hosts, the have a PowerCLI script which would evacuate VMs from a host, reboot, then iterate through the remaining hosts.

      Here’s a post I wrote on this a while back:

      Let me know if you have any other questions!


  • stevenpoitras

    What sort of RPO are you looking for?

    With native replication you can actually get a RPO of 15 minutes by having cascading replication schedules (e.g. 4 hourly replication schedules offset by 15 minutes)

  • Anil More

    Hi steve,
    May be its a stupid question to you but I need your a help on nutanix acropolis can we attach a disk to Vm using UUID ?If yes then How?

  • Jason Gersekowski

    Hi Steve,
    With the release of the “Volumes API” so that iSCSI can now be presented directly from the NDFS Storage, does Nutanix have plans to allow connectivity to this storage from servers that are not Intel based (eg. IBM AIX, HP HP-UX or Oracle Solaris server), to allow further consolidation of SAN resources ?



  • Pablo Scheri

    Hi Steve,

    Reading you bible, I realized that being a distributed all component has a Master/Controller -> Slave configuration, and there are a lot of them.

    I was wondering, how much extra computing overhead does the Master/Controller role adds to a CVM? and that made me think if is it possible to “suggest” the infrastructure to use a certain CVM as master for each distributed service.

    For example, this could be useful in the case where you have a Storage Only host, that only runs the Controller VM managing cold data, and all master roles overhead could be delegated to that CVM as not to overload the cluster compute intensive ones.

    Hope that makes sense.

    Thank you in advance!


  • Ian Elsinga

    Hi Steve,

    I understand that the Nutanix recommendation for SQL Server HA is to use AlwaysOn Availability Groups without using Failover Cluster Instances with shared storage. In our case, however, it is advantageous to use a combination of AAG and FCI and Microsoft indicates this is a valid scenario ( Effectively, some of our instances would be FCI instances.

    My questions is what would be the disadvantages of creating FCI instances with shared storage within the AAG from a Nutanix perspective? Are there any performance considerations?


  • Sean

    Hi there Steve,

    I’ve deployed two Nutanix CE appliances and have a basic ‘cluster’ setup. I needed to shutdown the whole lot, so I issued a ‘cluster stop’ command first, and ensured all services were at DOWN. Then I shutdown the CVMs individually, then the actual host appliances themselves last.

    After powering everything back up again, I logged into one CVM and issued a ‘cluster start’. Everything went to a status of ‘UP’ however my Web UI page gives me an “Oops – Server error” message on both of the CVM addresses (https://cvmIP or https://cvmIP:9440/console).

    Any ideas what would cause this?


  • Christoph Herdeg

    Hi Steve,

    thanks alot for your great NutanixBible, imho it’s a must-read for everybody interested in hyperconverged SDDC solutions! For me, being a VCAP-DCA and coming from the VMware side, it’s missing a few quite relevant topics, which I would very like to see for my own better understanding:

    – A detailed descrition of a Nutanix Cluster and its requirements (in comparison to, e.g. a vSphere HA/DRS-Cluster)
    – Nutanix’ Hypervisor integration and feature comparison sheet
    – The Book of vSphere needs more flesh…

    Best Regards,

  • Kevin

    Suggested update to the bible. You use CVM all over the first two chapters, but don’t define the Nutanix Controller VM (CVM) acronym until chapter III.

    • stevenpoitras


  • Mathias

    Hey Steve,

    I’m in a bit of a pinch here… I need to populate PD’s automaticly through MS Orcherstrtor. This isn’t all too hard but I need a way to evaluate the “fullness” of a PD since we can only have 50 vm’s per PD. So when a new machine gets created, I need a way to run through all the existing PD’s and count the VM’s they currenty have in them and then add the new VM to the PD which has the least VM’s in it.

    I can do this pretty easely from ncli:
    ncli pd ls | grep ” VM Name ” | wc -l

    But I need a way to do this from powershell… what am I overlooking?

    Kind regards,

  • Skelly

    Hi Steve,
    I am very new to the Nutanix world. Is there a script that would automate the deletion of VDI’s from the protection domain? Currently it is a manual process for us. Thank you.

  • Cynthia Rojas Haddad

    Hi Steve, this is a different kind of question… I’d like to discuss a Sr. SW Development Engineer (Cluster Resource Management) position in Sunnyvale. We are Springpath, a growing company providing an enterprise data platform solution and would like to hear your thoughts.

    Are you interested?


    Software Development Engineer – CRM

    Software Development Engineer (Cluster Resource Management)

    Springpath is the pioneer in hyperconvergence software, turning standard servers of choice into a single pool of compute and storage resources.

    We are actively hiring a Sr. Software Development Engineer (Cluster Resource Management). You will design, develop, test, and debug cluster-wide resource management capabilities in Springpath’s innovative distributed file system.

    Required Software Design experience
    Distributed Systems
    Cache Coherency
    High Performance Systems
    Quorum Protocols
    Desired Software Design experience
    ESXi, Hyper-V, or other Hypervisor

    Required Educational Qualifications and Experience
    Master’s degree in Computer Science or related field, and at least 3 years (or equivalent) of software development experience

    Cynthia RojasHaddad
    Sr. Talent Acquisition
    C: 408.634.0689
    e: crojashaddad AT springpathinc DOT com

  • Shailesh Shah

    Hi Steve,
    With AHV is link cloned VM’s possible (just like creating link clone on ESXi without vCenter)? Requirements are ability to create Win 7/win 10 workstations in workgroup mode from Master Template.

  • Kevin Penn

    Steve, hope you are keeping well
    I am trying to upgrade a AVH cluster – three node and keep running into this error –
    Not Enough space on /home/nutanix partition on 192.168.XXX.XXX. Available = 3.28496170044 GB : Expected = 3.48101806641 GB
    Cluster is currently under-replicated
    Any ideas as to what i should do?
    I have deleted a few log files

  • alh

    Hi Steve,

    First of all thank you for your “Virtualizing Splunk on Nutanix” paper as it is a great reference for us.

    We are currently trying to size / design a dedicated Splunk deployment on Nutanix and we do hesitate between having few Indexers on “storage-heavy nodes” or more Indexers (one on each node).

    However, there are several references in the text to “One Indexer per node” :

    “The following section covers the Splunk sizing and considerations for running Splunk on Nutanix.
    We assume a minimum of one Splunk indexer per Nutanix node.”

    “Note: Assumes one indexer per node”

    Is it a good practice / recommendation to have at least one Indexer per node ? If yes, why is it so ?

    Or maybe that choice was best fitting your testing needs somehow ?

    Thanks in advance for your feedback,

    Best regards,

  • Kevin Thorpe

    Hi Steve, since the author of the wonderful Nutanix Bible is offering to answer questions I think I’ll take advantage.

    We’re looking into using Spark in a small scale on Nutanix in containers. Yep, throwing everything into the pot. I’ve been asked to install HDFS on the cluster but the thought of adding a resilient distributed filesystem on top of a resilient distributed filesystem doesn’t feel right. What I really want is a shared volume that presents on each Nutanix node as local. We use AHV so iSCSI and this is exactly how that works. The iSCSI device can be mapped from the local CVM and Stargate and the underlying block storage sorts everything out for us. I’d like to do this but sharing the volume. I know that’s not going to fly in iSCSI but I do know that ESXi uses NFS to access disk images and the old way to import those was to NFS mount the container and copy files in. Would NFS mounting the container from the local CVM for each Spark worker be feasible? I know I could set up a single NFS server to share this but that doesn’t allow Nutanix to localise the data and forces network traffic at all times.

    • stevenpoitras

      Hey Kevin —

      What you’re talking about is definitely possible, however might not be supported or recommended depending on the size and quantity of the files. For Spark, it’s more than likely going to be a great deal of files.

      One option could be to use AFS to act as a SMB/NFS share, another option is we also just announced an object store service (similar to AFS), that exposes storage via an s3 interface.

      What would you be using this for? Temp/ephemeral storage?

      • Kevin Thorpe

        Thanks for the quick reply. To be honest I think I’m barking up the wrong tree. I’m under pressure to install a full Hadoop cluster and I think to get all the things we want I’ll have to do that and manage it going forward. At the moment this is for one ETL process which manipulates and ingests about 300GB into an analytics db. I was leaning towards NFS but the S3 type store sounds perfect for this use case. The piece missing is data locality which I’m starting to think is going to be nigh on impossible if we’re running Spark in containers on top of VMs on Nutanix. The two concepts are diametrically opposite. I think I’ll need to have HDFS running in containers as well and use the Spark-on-Kubernetes modifications to work out the locality.

  • Mark Werner

    Hi Steve, Seems the links to the images in your Microsoft Failover Cluster Configuration with Nutanix post are broken.

  • Mx CA

    Dear Steven

    for one of our banking customer, we are proposing a Nutanix Infrastructure to run a Haddop cluster.

    Your representatives proposed an architecture similar to the one here depicted (NUTANIX RF2 and HADOOP RF2), although smaller (8 Nutanix nodes)

    However, a few questions come to mind:

    1) having RF2 both at Nutanix and Hadoop level, doesn’t this mean that a single block is effectively replicated 4 times ?

    2) as Nutanix cannot obviously control where Hadoop puts the data, is it possibile that both copies of a Hadoop block end up to the same physical node ? So, in practice, 4 copies of data do not necessarily mean RF=4

    3) wouldn’t be a better choice (in terms of HA and space efficiency) having RF=2 at Hadoop level with deduplication (that means RF=1) and RF=3 at Nutanix level ?

    Please let me know your opinion

    Best Regards
    MXCA – TIM Italy

    • Thad McCall

      My question is similar but with Splunk and Nutanix. I’d really like to user SPlunk indexer clustering to take advantage of search factor 2. But that means RF2 on Nutanix and RF2 on Splunk. That seems like a terrible waste of storage. Isn’t there a way to create a container with RF1 and let Splunk take care of replicating the data amoungst it’s indexers in the indexer cluster?

Legal Mumbo Jumbo

Copyright © Steven Poitras, The Nutanix Bible and, 2014. Unauthorized use and/or duplication of this material without express and written permission from this blog’s author and/or owner is strictly prohibited. Excerpts and links may be used, provided that full and clear credit is given to Steven Poitras and with appropriate and specific direction to the original content.