Bill Roth, Ulitzer Editor-at-Large

Bill Roth

Subscribe to Bill Roth: eMailAlertsEmail Alerts
Get Bill Roth via: homepageHomepage mobileMobile rssRSS facebookFacebook twitterTwitter linkedinLinkedIn

As many commentators have pointed out, VMworld is becoming the critical show for the IT industry.  Ironically, it occurred during Dreamforce which might be the other fastest growing show.

I think Nexentians really enjoyed our VMworld experience.  The attendees at the show are generally fairly technical.  Ask them about whether storage is too expensive on a per I/O basis, too difficult to use, or just too unsafe and they get it - immediately.  Plus they ask all sorts of smart questions.  Some of my favorite questions I heard when stopping by the booth included:

- Q: “What VAAI primatives do you support?”  A: we support all four, however currently we only support them for iSCSI.  NFS support is planned as well.

- Q: “Do you recommend the use of iSCSI or NFS for virtualization?”  A: we tend to answer this by saying either works.  I personally prefer NFS as it simplifies management and I think our global namespace cluster and NFS are hard or even impossible to beat.  However - KT runs us with iSCSI and, as our 3.1 release indicates, we continue to invest in iSCSI and block performance.

- Q: “What is the performance you are getting when running the Hands on Lab?”  A: Look here.   We were running a real time feed of performance that we call the Aura interface.

Aura will be making its way into NexentaStor soon.  It leverages the amazing data provided by DTrace.  One reason we were not too concerned about showing this pre-released feature to our major competitors by showing it at VMworld is that no other major storage company has the level of performance granularity DTrace gives us.  They may be able to emulate / rip-off the cool interface, but to match the quality of the information will require cracking open their kernel.

Somehow I think legacy storage has higher priorities, like fully supporting multi-core processors and the upcoming Sandy Bridge.  Or building in-line dedupe.  Or adjusting to the post hardware RAID world by delivering air tight data integrity and 256-bit checksums.  Or, for that matter, offering infinite linked clones - great for VMware.  All of these changes require changes to the fundamental architecture of their kernel and file systems, which is why it is proving so hard for legacy storage to catch up to us.  They’re 32bit and we’re 128bit and that’s the way it is going to stay for a long time 

In any case, here is the Aura interface that helped so many attendees understand the storage performance required by Hands on Lab:

Starting in the lower left and moving clockwise, this interface shows:

- Most frequently accessed files.  In the case of the Hands on Lab, this showed which VMDKs were being accessed the most.

- Backend IOPS and then NFS operations.  Pretty self explanatory.  Yes, what data is presented is configurable.

- The circular area: the outer ring shows reads in blue and writes in green from each ESX server running the labs.  The chords attaching these servers to NexentaStor, which is in orange, show the bandwidth utilized.  The darker the color the greater the bandwidth.

The totals for the show are quite impressive - well over 2 billion operations were performed by the four NexentaStor instances running.  The peak IOPS per system were several hundred thousand IOPS.  Over 200,000 VMs were created and destroyed during the show.

And, here’s the kicker, NexentaStor was running on industry standard hardware from Supermicro with STEC drives for write and read cache and 7200 rpm SAS drives for capacity.  Monday some DRAM on one of the four servers (two HA pairs) failed.  And no end users noticed because of our HA cluster performed correctly and failed over.  Meanwhile our load increased from a designed 33% to over 60% of the total load of the Hands on Lab due to unspecified issues with either NetApp or EMC.  These vendors shifted to very heavy DRAM and flash based systems, and one vendor resorted to an a several million dollar DRAM and flash based array to keep up with the performance load.  We estimate that our implementation was over 90% less expensive than the other solutions running and yet it was at least as fast and we believe it was the only system not to miss a beat as the load increased dramatically during the course of the show.

This blog is getting pretty long.  So I’ll just mention a few other highlights of the show:

1. VDI in a box: Racktop Systems was demonstrating their VDI in a box solution, which they call the vBOX.  This kind of system is exactly what the market needs IMHO.  We have lots of happy VDI customers and are fast becoming the VDI solution of choice for both Citrix and VMware deployments.

Typically buyers and analysts focus on NexentaStor as providing IOPS and enterprise class reliability and data protection without the absurd pricing of legacy storage.  That’s true.  The other benefit, though, that various VMware S/Es and field engineers are telling us is the ability to scale the system as the VDI deployment progresses.  This ability to scale an existing deployment is radically different than the legacy model, where if you start with a Chevrolet you simply cannot scale it up in terms of performance to the GMC loads required; to strain the analogy a little further, you have to start with a massive GMC if you believe you’ll eventually need the load.  By comparison, you can start with a vBOX from Racktop with 20,000 IOPS and scale that to 100,000+ IOPS and then join that together into a massive name space cluster and so forth.  This means you are not risking a multi-million dollar storage expenditure day one of the VDI deployment.

Racktop has shown that they understand the VDI use case extremely well.  By including VMware View and NexentaStor in one box Racktop delivers a turn-key solution that can handle the random I/O and unique I/O patterns of VDI.

2. VDI at the Intel booth: We had a great experience in the Intel booth, demonstrating breakthrough cost-performance of VDI as measured by the Login VSI benchmarking solution.  Many thanks to Marco Righini from Intel who is one of the industry’s leading minds in how to implement and operate VDI systems; plus Marco is a great pleasure to work with.   The Intel booth, not surprisingly, was in a great location right across from VMware.

3. One PB for $475,000 list price with HA: Aberdeen is a well regarded VMware expert with a long history in storage and with NexentaStor in particular.  They chose a very direct way to make a splash at VMworld.  They ran, in their booth, a full 1PB of enterprise class storage, with HA, full replication, VAAI integration, and the opportunity for SSD based cache and they put a price on it.  Only in storage would the fact that they’ll actually price it out and stick by that price make news - but before Nexenta broke onto the scene as the leader of OpenStorage such behavior was unheard of!   Ask your legacy vendor to tell you right now how much 1PB would cost.  Their answer will be: “it depends.”  What does it depend on?  The price you pays depends on a complicated algorithm that includes your negotiating ability and their need for a particularly high margin deal at that point in time.

Back to the headline pricing - $475k for 1PB works out to $475 per TB, or approximately 93% less than the list price of legacy vendors.  And yet Aberdeen’s solution was based on NexentaStor which, as discussed above, ran VMworld’s Hands on Lab and proved itself to again be able to perform as well as legacy storage while maintaining at least as high a level of availability.

So those are a few highlights of the show at least as I see it from a very Nexenta-centric perspective.  What did I miss?  Let us know your thoughts.  See you in Copenhagen.  And Nexenta will be at various shows including this week’s Phoenix VMug, along with our excellent partners Nordisk.  Please stop by.

Read the original blog entry...

More Stories By Bill Roth

Bill Roth is a Silicon Valley veteran with over 20 years in the industry. He has played numerous product marketing, product management and engineering roles at companies like BEA, Sun, Morgan Stanley, and EBay Enterprise. He was recently named one of the World's 30 Most Influential Cloud Bloggers.