Original URL: http://www.theregister.co.uk/2010/08/31/emc_specsfs2008/

EMC embraces benchmarks

Celerra gateway excels

By Chris Mellor

Posted in Storage, 31st August 2010 16:55 GMT

EMC has started getting enthusiastic about benchmarks, the latest being SPECsfs2008-nfs v3. It submitted a Celerra VG8 Gateway system - that's a NAS head sitting atop either Symmetrix VMAX or CLARiiON Fibre Channel storage. This VG8 is a bit of a monster with up to eight data movers, called X-Blades, which contain 6-core Xeon 5600 CPUs cycling at 2.83GHz.

The benchmark system had just one of these X-Blades with a second standby one, and recorded 135,521 SPECsfs2008 operations per second with an overall response time (ORT) of 1.92 sec. EMC says this is a hardcore result as each Westmere core is running at 22,500 operations per second and that's probably a higher IOPS total per core than any other submitted system.

The highest recorded SPECsfs2008 result is 333,574IOPS with an ORT of 1.68msecs from a HP BL860c i2 4-node cluster using blades with two 4-core Itanium 9300 processors; not directly comparable on a core basis.

A Huawei Symantec N8500 Clustered NAS system delivered 176,728IOPS with a 1.67msec ORT. So we can see that the EMC VG8 result in itself is not that impressive. The multi-tiered Avere FXT 2500 NAS accelerator recorded 131,591 IOPS, with a 1.38msec ORT - in EMC's ball park.

NetAppp's best result is the FAS608 high-end product which reached 120,011 IOPS and an ORT of 1.95msec. Just on a IOPS basis this is no longer impressive.

If the VH8's performance scales linearly then one with four X-Blades would run at 542,084 SPECsfs2008 IOPS and a 7-bladed one, leaving a blade for standby, would run at 948,647 IOPS, with the 8-blade job surpassing a million SPECsfs2008 IOPS. We'll have to wait and see if EMC does submit multi-bladed VG8 systems to this benchmark. It provisionally looks as if it could have a record-breaking system on its hands, one that will give Avere, Symantec/Huawei, HP and everyone else something to think about.

Hint: suppose we use 6-core Westmere Xeons too? ®