EMC kills SPEC benchmark with all-flash VNX
EMC has topped the SPEC NFS benchmark rankings, scoring 497,632 operations/sec, using virtually all-flash VNX arrays.
The previous top SPECsfs2008 NFS v3 score was 403,326 ops/sec from an IBM SONAS (Scale-Out NAS) system using 1,975 disk drives. There were 1,680 x 600GB and 240 x 450GB SAS hard drives and a total exported capacity of 903.8TB.
EMC's configuration had 436 x 200GB solid state drives (SSDs) and 21 x 300GB SAS hard disk drives, in four VNX5700 enclosures, with a total exported capacity of 60.243TB. The difference that flash makes is substantial; you need far fewer drives and less capacity than with spinning disks to get the SPEC ops/sec numbers up.
The latest EMC result compares to a previous EMC Celerra VG8/Symmetrix VMAX benchmark, which achieved 135,521 SPECsfs2008 ops/sec using 512 disk drives and a total exported capacity of 19.2TB. The VMAX can support many more hosts than the VNX arrays of course. How would an all-flash VMAX rate on this benchmark?
The VG8/VNX configuration is a corner case, a red-line configuration set up to see just how fast VG8/VNX would go. It is a screamer but unrealistic – how many customers are going to populate four VNX5700s with flash drives in real life? It would be vastly expensive. However it is a marker for the future, showing that flash drives will trounce hard disk drive-based storage in I/O-intensive benchmarks.
We understand that EMC will likely announce even more impressive CIFS numbers in a couple of weeks time, "way bigger" according to one person familiar with the situation.
What of EMC rival NetApp?
A FAS6240 SPECsfs2008 benchmark of that system delivered 190,675 SPECsfs2008 ops/sec, using 288 hard disk drives with a total exported capacity of 85.8TB. That was just a few months ago, yet the system now looks seriously underpowered in top line number terms.
HP scored 333,574 ops/sec on this benchmark with a BL860c i2 4-node cluster using 1,480 hard disk drives and a 51.4TB total exported capacity.
The SPECsfs2008 benchmark doesn't reveal cost numbers, so we can't compare these systems on a $/SPEC ops measure. Get the flash-based systems ops/sec number high enough though, and the $/op number could be cheaper than the $/op amount of a disk-drive-based system.
We might speculate that comparing what is virtually a flash storage array with a disk storage array on the same benchmark is an unfair exercise, apples meeting oranges, that sort of thing. Such a tactic won't cut much ice with customers focused on top line numbers. They will look at these almost absurdly high numbers and see headroom, scalability for growth room, if they buy a smaller configuration. It is a halo effect.
It is obvious that IBM could, if it wished, revisit this benchmark with a flash-based SONAS system. Similarly, NetApp could test a flash-based 6240, and ditto HP. But they would each have to surpass half a million ops/sec for the exercise to be worthwhile. For all three companies, any ideas they previously had of having top-level SPECsfs2008 results using disk drives have been blown out of the water by this EMC result. It is a watershed benchmark moment. ®