Larry Ellison founded Pillar Data on the premise it would become the next EMC or NetApp, and go through an IPO or be bought. On its way there it sees a new system that could replace 100 NetApp filers coming by 2012. This perspective on Pillar was given by its CEO and CFO at a customer summit in Phoenix, Arizona on March 10. CFO …
apples to apples
3PAR pulls of 224k IOPS with 1280 disks not 2000. It was IBM that posted a ~360k IOPS number with a pair of arrays with 2000 spindles.(and a $7M price tag)
And I'm sure since they are taking the next 18 months to develop this new thing by the time it comes out the other competing vendors will have similar systems as well. SSD took the entire enterprise industry by surprise and now everyone is scrambling to come up with suitable technology to best leverage it, nothing on the market in my opinion can do it effectively at this point outside systems from TMS (which I've never used, so only basing this on their on paper numbers).
The Axion 600 SPC-1 numbers posted last year were not impressive, especially on the cost front. Not only did it not perform very well but they had to short stroke the drives in order to get the performance that they did. For slightly less $$ you can get nearly 3x the usable storage capacity, and an extra 30,000 IOPS with a 3PAR F400 vs an Axiom 600. Now I don't know how Pillar licenses their software, 3PAR does charge extra for their stuff, but I think in the end it's still more cost effective and higher performing. And 3PAR has significantly more features, pretty much all of which are run in silicon, not software.
Fortunately these days new SPC-1 numbers include a % utilization number for usable storage to call out those who short stroke their disks.
The only thing I see Pillar having on 3PAR is Pillar can address more raw storage in a single system than 3PAR, something 3PAR better be fixing soon(my biggest gripe on their stuff). I find it kind of funny if you want to achieve high performance and high capacity(1PB+) with 3PAR you need 2 arrays, to address the capacity issue, if you want the same from the competition you need 2 arrays(or more), to address the performance issue.
Actually as far as features go it seems Pillar is light on those, not surprising though all of the smaller vendors(3PAR included) typically don't the 10,000 strong army of developers making new things so they have to focus more and do better at what they DO do.
The strategy is to copy Oracle Sun 7000?
Hmmm...use industry standard parts instead of proprietary, for unified storage requirements? Oracle is already doing this with Amber Road, aka the Sun 7000 series, aka the ZFS Storage Appliance. The lines of differentiation between Pillar storage portfolio and Oracle's strategic storage portfolio via Sun are diminishing, not widening.
Now that's what I call copy
"To cope with those you need the 800lb gorilla canine and Workman hopes that Paso and Napa are two dogs that will worry the heck out of the storage supplier sheep out there"
After writing that one, you've just got to put your feet up on the desk with your hands behind your head and bask.....
Pillar Data = 800lb albatross
The Pillar architecture is ridiculously inefficient -- here in 2010 a measly 226 IOPS per spindle at 22 milliseconds response time (SPC-1) is really awful. That Pillar had to short stroke the disks to get even this level of crappy performance is absurd.
Consider, IBM achieved 224 IOPS/spindle in 2004 with SVC version 1.2 -- using Intel 1U off-the-shelf servers and back-end "Bricks" from LSI Logic (DS4000 -- 2002 vintage).
Consider further...LSI's current array hardware (Sun 6180/IBMDS5020) does 330 IOPS per spindle (50% more IOPS) and does them at 11ms. (twice as fast as Pillar) -- and does them USING A FULL DISK STROKE.
And oh-by-the-way, the LSI/Sun 6180 costs half as much per IOPS and half as much per GByte as Pillar.
Put another way...Sun's LSI-based system can deliver equal capacity, IOPS and response time using about half as many spindles as Pillar -- so there's the green aspect.
Pillar's performance efficiency is bad because it's architecture is fundamentally flawed. Eight years and half a billion dollars of Larry Ellison's money have resulted in what is arguably the worst performing architecture in it's class. Pillar's miniscule market share reflects this.
From the CEO's comments, it appears that Pillar is looking to "fix" it's architecture at both the low and high end primarily by replacing HDDs with SSDs.
Do Pillar and Oracle actually believe people are going to put expensive SSDs behind the least efficient SAN architecture on the market?
Will customers be amenable to paying huge dollars (and fat SSD gross margins) to Pillar for the privledge of plugging them in behind Axiom -- then losing 40% of their IOPS and adding several milliseconds of response time in the process?
Not bloody likely.
Dear Anonymous Coward
Oh for god’s sake! Instead of flip-flopping between the IBM DS 5020 results and the Sun 6180 results to present the best numbers, I’d suggest looking at the data from a single storage array perspective.
If you review the IBM DS5020 results, the revised (revised on March 11, 2010) SPC-1 Price-performance numbers indicates $8.46/SPC-1 IOPS. (http://www.storageperformance.org/benchmark_results_files/SPC-1/IBM/A00081_IBM_DS5020-Express/a00081_IBM_DS5020-Express_SPC1_executive-summary-r2.pdf).
If you review the Sun 6180 results from October 2009, the SPC-1 Price-Performance numbers indicate $4.70/SPC-1 IOPS. However, the price used in this calculation uses a 38% discount! Now, if you don’t discount the system (like the IBM DS 5020 results), you get $7.60/SPC-I IOPS. (http://www.storageperformance.org/benchmark_results_files/SPC-1/Sun/A00084_Sun_6180/a00084_Sun_6180_executive-summary.pdf)
The Axiom SPC-1 Price-performance number is $8.79/SPC-1 IOPS from results that were generated in January 2009. Since then, Pillar has released Slammers (storage enclosures) that have improved its performance by over 50%. Additionally, the price of the configuration that is listed has been changed significantly in favor of a lower Price-Performance number.
If my math is correct the IBM DS5020 would cost $18.80/GB and the Axiom 600 would cost $13.50/GB for the physical storage capacity of the configuration. Therefore, I am not sure how you inferred that the $/GB for the IBM DS5020 was half of that of the Axiom. Pillar also offers an 80% utilization guarantee without any performance degradation of the system. This is done by providing unique quality of service for the various LUNs and FileSystems (Yes – the Axiom supports NAS and SAN from a single virtualized storage pool) based on priority and access patterns.
Another thing to note is that the IBM and Sun storage systems which are OEM’ed from LSI uses 8Gbps FC front-end connections, whilst Pillar’s Axiom used 4Gbps FC front-end connections.
The Axiom Storage System (unlike the IBM DS5020 and Sun 6180) was architected with the intent of providing storage in a multi-tenant environment. The SPC-1 performance test provides a single workload and does not provide a varying workload like a multi-application environment. Moreover, the offerings from IBM and Sun would get pegged on performance if other storage operations such as rebuilds, snapshots and remote replication were to take place during the SPC-1 tests. The Axiom has a distributed intelligent architecture, which allows the system to isolate rebuilds to a storage enclosure and utilizes the Cache and CPU of the Slammers to perform quick operations.