Nutanix

Discussion in 'Business & Enterprise Computing' started by GooSE, Oct 4, 2013.

  1. tullytyro

    tullytyro Member

    Joined:
    May 31, 2002
    Messages:
    28
    Ok some great questions...and sorry this post is long.

    The key is that Nutanix is a software company - the intelligence is entirely in software (inside a VM running on each and every node - also making us hypervisor agnostic). Fundamentally, Nutanix believes :

    1. The convergence of Compute and Storage tiers makes sense (especially around performance and simplicity of administration - no more 'islands' of storage and blades etc). Sure, the Nutanix software in the 'controller VM' uses some of the compute resource, but at least now that same resource can be shared between infrastructure and workload VMs. The same reason why "compute" virtualisation became popular (using excess CPU to share amongst apps/servers at the same time) ... now you can do the same to the storage tier.

    2. Highly distributed share-nothing software model makes sense (ie. each 'node' is independent of others, but they all work together)

    3. Commodity hardware wins, Ethernet wins. Let the software 'work around' failures if/when they occur - who cares about the hardware in that sense. A Google data centre doesn't care about node failures and nor should yours..

    4. A pure scale-out architecture makes sense - start small and grow only when you need to.

    Fundamentally, since our founders came from Google, they've taken this google-like scale-out model and brought it to enterprise virtualisation...with no need for a "traditional" or dare I say it "legacy" SAN.

    If you have the time, watch this (16 mins) : http://www.youtube.com/watch?v=XG81gi4pTI4

    If you have less time, watch this (5 mins) : http://youtu.be/nSqwAxhFpA8

    If you are really buggered for time, watch this (110 seconds!): http://youtu.be/FYF234Bx3Pw

    Come on - you can spare at least 110 seconds right??! It could change your life.... :tongue:

    If you prefer to read, here is a blog post from our CEO: http://www.nutanix.com/blog/2013/06/01/software-defined-storage-our-take/

    In that post he also explains why we bundle hardware with our smart software too for those wondering.

    Here's a pic of what makes up a Nutanix "2U block" of 4 "nodes":

    [​IMG]

    Note that there is no 'back plane' here - each node communicates to other nodes (no matter how many) via Ethernet - normally via top-of-rack switches.

    Yes you need 3 Nutanix nodes as a minimum config in the 2U chassis. Once you start with 3, you can add extra nodes one at a time, forever. Remember each time you add a node, you are essentially getting another storage controller (which is in software!). Have 50 nodes? You now have 50 controllers...all working together and making your life easy. Go and have a beer instead of worrying about LUNs, Volumes, growth, outages, performance drops etc.

    This concept of software controllers also unlocks the power of "data locality" - essentially the data associated with your VM now 'follows' the VM around the cluster if the VM moves - because the controller VMs are now aware of both that compute and storage tiers...This is a critical point. Josh does a good job of explaining it here: http://www.joshodgers.com/2013/09/19/data-locality-why-is-important-for-vsphere-drs-clusters/

    This 'data locality' also ensures performance at scale remains as good as it was on day 1. See some slides here: http://imgur.com/a/pPheb


    In terms of usable space, I mentioned it on page 1 of this thread, but essentially its about 1/2 the formatted capacity of the disks available to the entire cluster (and that is because there is a 2nd copy of all data in the cluster - so if you write 1MB of data, a 1MB copy is stored elsewhere in the cluster for a total of 2MB 'used'), minus some for our software of course.

    Want more technical detail check Steve Poitras' page here: http://stevenpoitras.com/the-nutanix-bible/

    I just did a deployment today and the customer did most of the work and it took about 45 minutes (from him never seeing it before). There's videos online where people can do it in under 15 mins. At the end there were 4 ESXi hosts each 'seeing' a NFS datastore which is ready for VMs - all in a 2U appliance. Happy days. You could have a complete virtualisation project ordered and deployed within a few weeks (and most of that time is for the block to clear customs!).

    The great thing also about being 'software-defined' is that improvements in features or optimisation of existing features are delivered by software upgrades (with no cluster downtime too)... ie. in some upgrades we have delivered increased IOPS and reduced latency as the engineering gurus improve the code...all on the existing Nutanix hardware nodes... Cool.

    Nutanix will be at vForum in Sydney later this month, please drop by if you want to see a Nutanix block in person too and pick up a "No SAN" beer coaster. Shiny.

    Again, some great questions here. Good stuff.

    Forgot to add:

    Backups: Just use whatever you use now for VMs generally (assuming it is an IP based solution). eg. Veeam is a good example but there are others.

    RAID? No sir. The controller VM simply formats the disks. The controller VMs are in charge of replicating user data and re-replicating it when a new disk is put in for example (and formatting it) after a failure - so the process is quick. Re-replication of data happens automatically, so in fact N+1 can be restored automatically if you have enough nodes. Again, all this is based on how the big web boys like google / facebook etc do it.
     
    Last edited by a moderator: Dec 15, 2013
  2. DRAGONKZ

    DRAGONKZ Member

    Joined:
    Jul 3, 2001
    Messages:
    1,796
    Location:
    Bankstown, Sydney
    I've noticed the SMB 3.0 (Hyper-V 2012) support is finally getting closer.

    Have they added in the "Rack awareness" coding yet?

    ie, you have 2 x blocks in Rack A, and 2 x blocks in Rack B, and it can be configured so that the copies of each bit of data don't live in the same rack as where it originated from?

    We were told about 3 months ago that this was a feature that would be in the next release.

    Without trying to dull the product, we found that if you are catering for a big CIFS/SMB workload, then the Nutanix doesn't end up being as cost effective.

    Throwing $70K odd at a 6XX0 block for 14TB odd usable space doesn't go as far as it does throwing that $70K odd as some additional disk shelves for a traditional SAN array.

    Another issue that we started to see was that when very large workloads were placed on nodes, the controller VMs ended up ideally needing dedicated/reserved CPU/RAM resources. This obviously means less resources are now available to all your guests.

    You can obviously "add another node" to the solution, but this may not work for everyone's budget.

    I can't comment on after sales service, but can say that we were very impressed with the pre sales support.

    Some of the questions/concerns we had were resolved with several conference calls with Nutanix high level tech guys around the globe, and during out POC phase we had several Nutanix guys on site for weeks on end (or literally on stand by waiting for a call to be there).
     
  3. KiwiPete

    KiwiPete New Member

    Joined:
    Oct 11, 2013
    Messages:
    4
    Hi all,

    I am a customer of Nutanix based in NZ. The product is fantastic, the people are brilliant and the solution works. We had a "traditional" SAN / Blade environment that was only six months old that failed to meet the VDI project we were running with. Key points for us:

    Nutanix out performs our EVA P6500 SANs
    We save 40% of our CapEx budget
    We are running 100% (200 virtual desktops) VDI and production servers on Nutanix kit.
    Deployment - we did it ourselves!

    Feel free to email virtual(at)stdc.govt.nz for further discussion.
    We've been on Nutanix since December 2012 and will not be going back to a SAN based solution.
    Pete - IT Manager (STDC)

    The "Block" (chassis) is free - you buy nodes and when you run out of space in the Block, they send you another block. We run 10 nodes (6 in production and 4 in DR). Cheers
    Pete
     
    Last edited by a moderator: Dec 15, 2013
  4. ebar

    ebar Member

    Joined:
    Oct 5, 2013
    Messages:
    15
    I am not aware of any product that can solve every need for every customer.
    Your feedback is interesting as it mentioned other type workloads, CIFS/SMB which proves that cost always "interfers" with technical opinions :)

    When it comes to performance my experience is that you can never get rid of a bottleneck, its just keeps being moved around. IE, if you were to add RAM to a VM then you might have moved the bottleneck to the disks etc. So its important to know your workload and try to place the bottleneck where its least impacting you.

    I myself is NOT a fan of SANs, far from it. But they do serve their purpose. I too think NAS/IP storage is better and scales better (after all isnt the internet a form of NAS?)

    I think one of the issues here is capacity, raw vs useable. If you buy 20TB you ll get disks that are right sized for starters. IE a 2TB disk is in fact 1.67TB. Then you format it, lose maybe a little bit. and then in the case of Nutanix you invest 1/2 in a copy in the cluster. So going from 20TB raw I expect to get around 8TB useable? Would that be right?

    Note I said INVEST, not lose, to a copy. Everybody holds a copy of their data whether its DR or RAID etc. So this is industry standard.

    I welcome Nutanix and will keep an eye on them.
     
  5. OP
    OP
    GooSE

    GooSE New Member

    Joined:
    Jun 26, 2001
    Messages:
    6,679
    Location:
    Sydney
    One of the things that worries me a little is what if you need more storage but not more compute? You buy another node and have CPU/RAM working at close to zero capacity just for a few more TB. Seems a little wasteful.
     
  6. tullytyro

    tullytyro Member

    Joined:
    May 31, 2002
    Messages:
    28
    The new 6020 series is designed just for this purpose - low spec cpu/ram just to run a the software controller to allow for storage expansion only. Of course you can order it with high amount of ram so you can run normal workloads on it too - your choice.
     
  7. ebar

    ebar Member

    Joined:
    Oct 5, 2013
    Messages:
    15
    that was the issue we had with the IBM XIV as well. But the thing is over time everybody needs more of everything, CPU as well.
     
  8. tullytyro

    tullytyro Member

    Joined:
    May 31, 2002
    Messages:
    28
    Not yet - stay tuned though. Hyper-V is in tech-preview stage right now.

    Fantastic point.
     
    Last edited by a moderator: Dec 15, 2013
  9. ebar

    ebar Member

    Joined:
    Oct 5, 2013
    Messages:
    15
    As valid as that point is mate, consider this: some customers will not need 1 copy in primary site and another one in DR. For those customers it means you have 1 copy too many. Those customers will typically not require high IOPS/throughput. Think CIFS/SMB/office style workload.

    Still in some instances where you can avoid going into DR mode (which typically entails performance hits as distance/latency/network can become an issue) it would be preferred to not have to go across the wire to DR site = better ROI. For those customers I reckon having 1 copy in primary site is an asset. I suspect this type of customer is the target for Nutanix?
     
  10. tullytyro

    tullytyro Member

    Joined:
    May 31, 2002
    Messages:
    28
    True, and because Nutanix is software-defined, you can actually change the number of copies.... By default it is set at 2, but some use cases it may make sense to only set it to 1 (ie. no replica copy, eg just big data analysis projects) or 3 (extra paranoid, at the "cost/investment" of more storage space in the cluster).

    But for the reasons already mentioned in this thread, the majority of people like having the 2nd copy 'ready to go' next to the primary in case of node failure (for example).

    I'm not saying Nutanix is a perfect fit now for every single use case in existence but we are only just starting this journey into a software-defined world and true convergence. But if you have a performance problem which you think may be due to poorly performing SANs or latency-sensitive apps which may benefit from the converged architecture then it is worth taking a look.

    VDI is a classic use case, but we are seeing more and more general server workloads being deployed on Nutanix as well of course.

    The support team is amazing IMHO, as are the engineering teams. Nutanix had engineers join from Facebook and Google to develop new features and improve the current offerings (eg dedupe is in the latest software) - it's always fun on the 1st of the month to see who has joined the company and where they've come from (some even come from the SAN vendor land).

    We also have 4 VCDX's now and I think more are joining soon; so we have a lot of VMware expertise in-house. A lot of these guys are working on reference architectures etc for the most common server workloads for example.

    We are always on the lookout for stellar talent - so check the careers page if you'd like to join the ride on the Nutanix web site.
     
    Last edited by a moderator: Dec 15, 2013
  11. ebar

    ebar Member

    Joined:
    Oct 5, 2013
    Messages:
    15
    Thats pretty awesome, cant not like that ey :)

    I did get in touch with Nutanix, had a couple of phone calls. Was found too light, which is OK of course. However when I asked HR where I fell short (so I can upskill) I never heard back. That is a bit disappointing as I dont know what to fix. :confused:

    Im off the market now and start a new role in Melb. on Monday.
    I think Nutanix is on quite some ride, enjoy it!
     
  12. DRAGONKZ

    DRAGONKZ Member

    Joined:
    Jul 3, 2001
    Messages:
    1,796
    Location:
    Bankstown, Sydney
    Several people at my company (including myself) spoke to you as one of the references we needed when we were doing our POC.

    It's a small world :lol:

    Did you guys end up moving your back end infrastructure over yet, or are you still just running VDI on it?
     
  13. KiwiPete

    KiwiPete New Member

    Joined:
    Oct 11, 2013
    Messages:
    4
    Hi!,

    We've migrated key production servers and all desktops to the new Nutanix kit. Just loving the performance. We still have the EVA6500 sitting in the background with some production servers on - just a matter of time to migrate those across - no rush. 6 nodes of 3000 in productions, 4 of the 2000 series. So we are running 200 VDI and about 8 Production servers. Southland District Council here in NZ and I think Hornsby Shire Council in your neck of the woods are / or will be following our methodolgy.

    Now to be honest the reason for joining the forum is I'm into Star Citizen and while I've been in IT a long time, OC is a skill I never bothered to learn until now. But I'll never turn down an opportunity to talk about Nutanix or Unidesk!
     
  14. PabloEscobar

    PabloEscobar Member

    Joined:
    Jan 28, 2008
    Messages:
    13,835
    So what was the total cost of the infrastructure to Run 200 VDI desktops and 8 Production servers? Are you using Horizon View for your VDI?

    Just trying to get a baseline figure so I can do some rough comparisons between this, and a 'traditional' servers + san model.
     
  15. tullytyro

    tullytyro Member

    Joined:
    May 31, 2002
    Messages:
    28
    If it was the exact same price as a traditional server + storage model, which would you choose and why? :)
     
  16. Jase

    Jase Member

    Joined:
    Jun 28, 2001
    Messages:
    196
    Location:
    Sydney 2081
    Hey KiwiPete, (TullyTyro possibly),

    How do you judge that the performance was better? Is there specific performance metrics you compared?

    What about Nutanix do you think makes the performance better than a traditional SAN exactly? Was the EVA always a bit of a dog and you never understood why ?
     
  17. PabloEscobar

    PabloEscobar Member

    Joined:
    Jan 28, 2008
    Messages:
    13,835
    I guess it would depend on the workload and future direction of the company/environment I am working for.

    Are the prices exactly the same?
     
  18. fredhoon

    fredhoon Member

    Joined:
    Jun 27, 2003
    Messages:
    2,643
    Location:
    Brisbane
    Agreed, the conversation does read like a scripted Q&A.
     
  19. PabloEscobar

    PabloEscobar Member

    Joined:
    Jan 28, 2008
    Messages:
    13,835
    Member since 2002... Playing the Long game... I like it :).

    More likely is that Tulltyro shot an E-mail to one of his customers (kiwipete) who has agreed to be used as a case study asking him to stick his head into this thread and outline his experiences.
     
  20. tullytyro

    tullytyro Member

    Joined:
    May 31, 2002
    Messages:
    28
    Sorry for the late reply guys, was in Canberra today helping with the first hyper-v / nutanix deployment (will hypervisors become commodity along with the hardware ? Perhaps a discussion for another day...)

    KiwiPete is a customer and I encourage existing customers *and* potential customers to this thread because it is good to get the whole story and some of the questions have been really good. There are even people who tried Nutanix and then decided to stay with the traditional architecture on this thread. It's all good.

    KiwiPete did a webinar with Unidesk and Nutanix on his experiences and I think he goes through the numbers in this video and I think there was a slide on comparing the HP SAN performance to Nutanix: http://www.unidesk.com/videos/customer-webinar-south-taranaki (skip to 13min go get to his bit and ignore the sales bit at the start should you choose).

    If you are in Sydney next week please drop by vForum and speak to us - and Josh Odgers (VCDX) is doing a preso on Converged Infrastructure as well (and he will be on our stand). We will have a block there. Mention you saw this thread for a free gift!

    They certainly could be. 200 VDI users could be catered for a number of ways, depending on what is most important (less nodes/space therefore more density or scaling out the smaller/cheaper nodes but more of them and more rack space etc) and the specific use case. VDI is hard to size and scale (and maintain performance) - hence Nutanix makes that headache easy by the 'start small and scale' mantra.

    Lets say a node of a particular model might cost $20k. If you have 3 of them then it's $60k. If you have 10 of them it will be $200k.... but you can buy nodes (after the initial 3) one at a time forever - only when you need it.

    "Start small and scale" vs "buy big now and wait to fill it".

    Don't underestimate how easy Nutanix is to administer. If you know vCenter you know Nutanix.

    Nutanix ships with eval versions of the hypervisor you want - so normal software licensing is extra of course (as it is for normal servers anyway).
     
    Last edited by a moderator: Dec 15, 2013

Share This Page

Advertisement: