Hitachi Performance Leadership, The Teddy Roosevelt Way
by Bob Madaio on Oct 10, 2013
Of the many things US President Theodore Roosevelt is known for, one certainly is the quote “Speak softly and carry a big stick, you will go far” and made me think that Teddy would have made a great HDSer.
You see, while other vendors are loudly proclaiming performance leadership after upgrading systems that were sorely in need of a lift (yes, I’m looking at you EMC), Hitachi continues down our path of consistently providing our customers the most innovative hardware architectures around.
Only without all the yelling.
Why has this come to mind now? We’ve recently completed SPECsfs2008_nfs.v3 performance testing of two new Hitachi Unified Storage VM (HUS VM) configurations, both “all flash” leveraging our patented Hitachi Accelerated Flash storage and both with our newly announced Hitachi NAS technology.
And the results prove we have the speed to lea…. well, let’s just say they are pretty awesome.
Our new configurations were:
|Storage System||File Component||Flash||Posted Results|
|HUS VM||2-Node HNAS 4100 Cluster||32 Hitachi Accelerated Flash drives (1.6TB each)||Click Here.|
|HUS VM||4-Node HNAS 4100 Cluster||64 Hitachi Accelerated Flash drives (1.6TB each)||Click Here.|
The above configurations are nothing too extravagant from a hardware perspective: 2 and 4 node HNAS clusters and 64 or 32 flash modules. That simplicity is exactly why the results are all that much more exciting.
Our HUS VM 2-node HNAS 4100 system delivered 298,648 operations/second with an overall response time of 0.59 milliseconds. While both numbers are astounding for the amount of hardware deployed, note that the 0.59 milliseconds overall response time was the lowest reported on this benchmark. Ever.
Our HUS VM 4-node HNAS system delivered a whopping 607,647 operations/second with an overall response time of 0.89 milliseconds. Again, amazing results. But the throughput numbers start to get so large that it might help to understand them by looking for a relevant comparison.
The most timely, and arguably most relevant, comparison might be the recently and LOUDLY announced VNX8000 (result here)… king of the recent “VNX2” launch. It certainly showed up to this benchmarking match ready to rumble, as it had twice as many “X-blades” installed (eight) to drive NAS traffic as did our 4-Node HNAS configuration and five hundred and forty four (yes, 544) SSD drives compared to our seemingly impossibly outgunned 64 Hitachi Accelerated Flash Modules.
Despite the should-be insurmountable hardware advantage for the brand-spanking-new EMC system, it actually drove 5% LESS NFS operations per second than our significantly more efficient HUS VM configuration, with both systems providing sub-millisecond overall response times.
Granted, in the VNX architecture one X-blade needs to sit idle waiting for an issue to arise before providing value and EMC does not have advanced, enterprise flash capacity like our Accelerated Flash, but you’d expect it to be able to beat out a system with half the installed file nodes and less that 1/8 the amount of flash devices, wouldn’t you?
With that comparison helping set context, the next logical one might be to NAS market “leader” Netapp. The most relevant system that Netapp has published results for is the FAS6240 in a wide variety of cluster sizes. In all fairness, it becomes hard to make a logical comparison between the systems because the HNAS per-node performance is >2X that of a FAS6240 node and NetApp’s flash-strategy seems to lag its benchmarking so only eight FlashCache cards are leveraged in Netapp’s benchmark.
Thus, the closest comparison is probably an 8-node FAS6240 cluster (results here), but despite having twice as many file-serving nodes and 576 power-hungry disk drives it still provides 18% less operations per second and is unable to provide a sub-millisecond overall response time.
Of course, benchmarks are not the real world, though industry-trusted ones like those at SPEC.org do their best to maintain a useful level of openness and vendor comparison for end users. That ability to compare is important, because NAS solutions for such things as large scale VMware deployments and Oracle databases among (other use cases) continue to gain significant traction and demand extreme performance. Customers however do not want to simply throw massive amounts of filers, disk drives and SSDs at every problem, and they are realizing that the system architecture does, in fact, matter.
So we are rightly proud of our architectural advantages that allow customers to deploy more efficient solutions and get more predictable performance. Yes, the market is awash with hardware providers whose design point is much more about developing to the lowest possible cost, while putting the onus on customers to deploy more hardware to make up for architectural and design limitations.
We choose another path. We choose to develop better hardware to provide our customers with highly functional systems that provide predictable (and yes, best-in-class) performance in the most efficient way possible. Some might say that’s a harder path, and maybe they are right. But results like these, and more importantly the continued success of our customers, have us convinced it’s a better path.
To learn more about our architectural differences that enabled this success, here are some links for the technically inclined among you:
- Hu Yoshida on HUS VM architecture
- Hu Yoshida on Hitachi Accelerated Flash architecture
- Matthew O’Keefe on Hitachi NAS architecture: Part 1 and Part 2
So, while other vendors speak loudly, launch loudly and deploy over-sized solutions. We’ll walk softly and provide the best technology we can. I for one, think Teddy would be proud.
Comments (3 )
Great article! It should come at no surprise that HDS is offering top of the line performance. This is what we do. Thanks Bob!
Bob, great blog. So whilst we can safely say HDS continues to outperform, out architect and out humble EMC and NTAP, some would ask “so what”? In real terms waht we delvier with our architecture is a price per IO that is dramatically lower and an overall TCO drastically reduced from anything our competitors can field.
To a CIO who is constrained by lower budgets, burgeoning data management requirements and a Data Centre that needs to be bigger and bigger every day with more power and cooling (all those spinning disks require a lot of juice), this has to be a marvellous thing. As data sets change and we require more NAS to fulfill our data ingress and egress needs this must be music to his ears. Let’s find out how much he saves (and publicise the heck out of it!)
[...] Hitachi Performance Leadership, The Teddy Roosevelt Way (10 OCTOBER 2013, DATA CENTER ADVISORS BLOG, HDS.COM) [...]