A New World Record for XP!

[#1]

Yes, while the worlds top athletes were in Beijing striving for world records in their various sports HP's XP24000 was setting a new performance standard of its own.  That standard is XP's new record SPC-2 benchmark result of 8.7GB/s!  That surpasses the previous record by more than 20%, and trounces the next closest single array competitor by more than 2X.  The swimmers in Beijing didn't set their new records by that kind of margin!

XP didn't buy this record either.  At a price performance of $187/MB/s XP24000 produced its record at 2.5x better price performance than the previous record holder.  That previous record was set by an IBM system of 16 separate storage arrays with 1536 disks and a virtualization appliance.  A single XP24000 set the new record with just 265 disks.  That's a single, bullet proof XP24000 taking on a whole team of IBM products, and winning!  Record performance, superior price performance, and bullet proof availability.  What more evidence could you want that storage consolidation with the XP24000 is a good thing?

I'd also like to point out that XP didn't have to be tuned like crazy to get this result.  We did have to play around with the network and servers to get this result.  8.7GB/s is new territory for the SPC-2 benchmark.  It took some effort to get it working at this level while keeping the network and servers out of the way.  But with the array we pretty much just set it up and it worked.  That means you can expect these kind of results on your real workloads without a lot of esoteric tuning.  Pretty nice!

Now a few words about the biggest no show in the high end performance race, EMC.  EMC is the only major storage vendor to never publish an SPC number, despite our open challenge for them to do so.  They make plenty of performance leadership claims, but show no visible proof!  Instead all they provide is vague arguments about SPC being unrepresentative.  I think everybody realizes that if EMC could publish a leading number they'd do it in about a millisecond.  Instead they won't even get in the pool with the XP!  I think we can all figure out why.

Why is XP24000's new result important to a storage professional?  The SPC-2 benchmark is the storage industry's standard benchmark for sequential workloads like video on demand, disk-to-disk backup, and large database queries.  XP24000's results show the ability to serve >10,000 high quality video streams simultaneously.  They also show the capacity to back up 36TB/hour which would keep 42 of the industry's fastest LTO-4 tape drives busy simultaneously.  XP put up last year's top SPC-1 number for a single, HDD based array.  Now XP has the SPC-2 record.  Both results have been verified by an independent auditor and reviewed by our peers on the Storage Performance Council.  I think this makes it pretty clear that XP is the gold medalist in storage performance!

[#2]

Where can I read more about this test/record?

[#3]

Hi Jeterman5000,

A paper was just published that you can read.  It's at h71028.www7.hp.com/.../4AA2-1990ENW.pdf.  

Thanks for asking!

[#4]

I have a thought for you guys over at HP.  Why not pull an EMC stunt, like they did on the EVA disk group and its capacity utilization.  HP should go out and purchase your own Symmetrix array, configuring it just like the XP in the above benchmark, same cache/# of drives, and then run the same benchmark.  Utilize the same everything across the board, # of servers, network "tweaks", etc etc.

then show us all the results

[#5]

Thanks for the suggestion MikeB.  I have to admit that would be a lot of fun to do.  But at this point we're going to stand by the principle that testing like this needs independent objectivity to be meaningful.  If a vendor runs the benchmark and has it audited and reviewed by the SPC that's a pretty good measure.  You know every array is being shown in it's best light.  An independent tester can be an excellent method too.  Without one of those factors these public results just can't be trusted.  NetApp's SPC-1 publication of EMC's CX was pretty showy, but could hardly have been considered a meaningful result.  I think the spectacle Chuck Hollis initiated a couple of weeks ago in the blogoshpere showed that EMC's testing wasn't very objective either.  No matter how you look at it biased testing is biased.

At the same time I think the implication of EMC's lack of results tells more than our publishing a result for them could.  It doesn't take much thinking to realize that if EMC's DMX could keep up with the XP24000 they'd have already published an SPC number.  Their continued silence speaks volumes.