SCI 2012Jun23 Latest SPECsfs2008 NFS ORT and throughput ops/SSD results

In Avere, FAS/V6240, FXT3500, NetApp, NFS throughput, Oracle, ORT, SPECsfs, SPECsfs2008, Sun ZFS Storage 7420 by AdministratorLeave a Comment

We return now to file system performance and our analysis of the latest SPECsfs® 2008* benchmark results. There has been only one new NFS submissions since our March report, that being from Oracle on their Sun ZFS Storage 7420 appliance.  Once again, there were no new CIFS/SMB results.

Latest SPECsfs2008 results

Lower is better in Figure 1.  The new Oracle submission was not fast enough to reach the top 10 in NFS throughput operations per second.  However, in Overall Response Time (ORT), the new ZFS 7420 came in at a respectable #7 with a response time of ~1.3 msec.

SCISFS120623(001) (c) 2012 Silverton Consulting, Inc., All Rights Reserved

SCISFS120623(001) (c) 2012 Silverton Consulting, Inc., All Rights Reserved

Figure 1 Top 10 NFS ORT Results

Top 10 response times are hard to do without using Flash or SSDs and Oracle was no exception with over 4.5TB of flash. The top 3 systems here all had flash (with the EMC VNX VG8 using over 87TB).

But the 44-node Avere FXT 3500 (#3) with 800 GB and the single NetApp FAS6240 (#5) with 1TB submissions continue to exemplify of what can be done with modest amounts of flash, at least for response time.  Both submissions exhibited response times under 1.2 msec. And given that non-data activity represents a majority of SPECsfs2008 operations  (less than 30% of the operations are Read or Write data), great response times with this little flash are pretty impressive.

SCISFS120623(002) (c) 2012 Silverton Consulting, Inc., All Rights Reserved

SCISFS120623(002) (c) 2012 Silverton Consulting, Inc., All Rights Reserved

Figure 2 Top 10 NFS ops/sec per SSD or FlashCache card

Figure 2 represents our second attempt at analyzing storage system flash utilization in SPECsfs2008 NFS activity.  This iteration seems better than our previous version (see March report) that measured NFS ops normalized by GB of NAND storage used in a submission.  Moreover, Ops/SSD or FC more closely matches the NFS ops/disk spindle we use for disk only effectiveness.

However, all SPECsfs2008 NFS results that used flash also employed at least some real disk drives.  As a comparison, the median for the top 20 NFS ops/sec/spindle is ~490 (not shown here) while the median top 20 NFS ops/sec/SSD or FC is ~40K or about 82 times as much as the pure disk measure. As such, any metric that combined SSD-FC effectiveness with drive spindle use would need to span three orders of magnitude of actvity.

The 44-node Avere submission with 4-200GB SSDs came in first while the NetApp FAS6240 with a single 1TB FlashCache card finished at second place.  All the other NetApp submissions on Figure 2 used a lone 512GB FlashCache card per node.

Significance

None of the other NFS or CIFS/SMB Top 10 charts changed so look to prior analysis for those results.

We continue to search for a better way to measure storage system flash effectiveness.  Given this and prior analysis Avere’s use of flash as a backend journal (ZFS intent log) and NetApps use of flash as cache beat all other approaches. But only minimal data are available (~20 flash submissions to date) so it’s unclear whether other results such as using flash and disk with auto-tiering would improve on current results.  But auto-tiering has a significant time lag (12 hours or longer in some cases) so use in a two to three hour long SPECsfs2008 benchmark may show little advantage.

With our second generation measure of flash effectiveness, what remains is an approach to measure total device effectiveness that spans both disk and flash as well as incorporating DRAM caching.   But then this would recreate the ChampionsChart™ included in our latest NAS Buying Guide (use the QR code below left if interested).

This report was sent out to subscribers as part our free monthly e-newsletter service.  If you are interested in receiving this and future storage performance analyses along with recent product announcement summaries, please use the QR code below right to reach our newsletter signup page.

~~~~~

[This performance dispatch was originally sent out to our newsletter subscribers in June of 2012.  If you would like to receive this information via email please consider signing up for our free monthly newsletter (see subscription request, above right) and we will send our current issue along with download instructions for this and other reports.  

Also, if you need an even more, in-depth analysis of NAS  storage system functionality, features and our latest ChampionsChart™ performance results please see our NAS Buying Guide available on our website.]

Silverton Consulting, Inc. is a Storage, Strategy & Systems consulting services company, based in the USA offering products and services to the data storage community.

* SPECsfs2008 results from http://www.spec.org/sfs2008/results/



Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.