Hard on the heels of Cisco bragging about a flash-assisted VMmark win, HP has kicked it into touch with a better score using a flash-boosted ProLiant server.
VMware's VMmark benchmark measures how a server runs simultaneous VMware virtual machines (VMs) which, in turn, run typical business applications. The VMs are grouped into sets of eight, called tiles. Each tile has a score and these are aggregated into a finished number, such as 10@7 tiles, meaning that the score was 10 with seven tiles running simultaneously.
If two servers both achieve 10, but one at 7 tiles and the other at 10, then the lower tile server is recommended by VMware, as a higher tile count may suggest that the server's workload was not properly balanced. Otherwise higher scores generally go with higher tile counts.
Cisco's UCS B200 M3 blade server, a 2-socket machine, scored 11.3@10 tiles on the VMmark 2.1 benchmark, using a Violin 6000 flash array with 16TB of raw flash. That beat an HP ProLiant DL385p Gen 8 server which achieved 8.31@8 tiles.
But an HP ProLiant DL560 gen 8 server – a 4-processor 2-node system, fitted with 400GB SAS SSDs and 146GB 15K rpm SAS 2.5-inch disk drives, 28 disk/flash drives in total – scored 18.27@18 tiles. There were two external storage controllers, each with 3TB of SSDs, a third with the hard disk drives, and a fourth with no drives. The configuration details can be seen here (PDF).
HP provides comparative 4-processor, 2-node server scores for our (and its won) pleasure:
- Fujitsu Primergy RX600 did 18.20@18 tiles using 261 drves.
- Cisco's UCS C460 M2 did 18@18 tiles using 261 drives.
- A Dell PowerEdge R910 achieved 17.63@18 tiles with 229 drives.
It's quite obvious, isn't it: SSDs respond to random I/O requests much more quickly than spinning disk does. The HP server wasn't that much faster overall than the others, but it needed less rack space, power and cooling to do the job because it used a few solid state drives rather than relying on disk spindles to feed the servers' I/O demands. ®