Comments on: Deduplication – The NetApp Approach http://ctistrategy.com/2009/07/20/deduplication-netapp/?utm_source=rss&utm_medium=rss&utm_campaign=deduplication-netapp CTI Strategy Blog Fri, 26 Mar 2010 22:05:17 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: Jesse St. Laurent http://ctistrategy.com/2009/07/20/deduplication-netapp/comment-page-1/#comment-1810 Jesse St. Laurent Mon, 27 Jul 2009 14:00:04 +0000 http://ctistrategy.com/?p=289#comment-1810 Kevin, Stay tuned. We are working on a VMware boot storm test. The boot storm is a bit of a synthetic test as well, but more realistic than my 100% random 4KB read test. We are waiting for handful of new Intel Nehalem servers to arrive to start the test. The plan is to run NFS over 10GbE. In the test results here, the host response time (latency) actually went down by a little over 80% for the higher NFS op test. This is not surprising as even though the CPU load and NFS ops are up, all of those requests are coming from cache. Nothing improves performance like eliminating disk operations. ;) - Jesse Kevin,

Stay tuned. We are working on a VMware boot storm test. The boot storm is a bit of a synthetic test as well, but more realistic than my 100% random 4KB read test. We are waiting for handful of new Intel Nehalem servers to arrive to start the test. The plan is to run NFS over 10GbE.

In the test results here, the host response time (latency) actually went down by a little over 80% for the higher NFS op test. This is not surprising as even though the CPU load and NFS ops are up, all of those requests are coming from cache. Nothing improves performance like eliminating disk operations. ;)

– Jesse

]]> By: Kevin Davis http://ctistrategy.com/2009/07/20/deduplication-netapp/comment-page-1/#comment-1754 Kevin Davis Fri, 24 Jul 2009 19:22:17 +0000 http://ctistrategy.com/?p=289#comment-1754 Granted...there's a lot of other things taxing the filer where we have the VMWare data housed. <a href="#comment-1753" rel="nofollow">@Kevin Davis </a> Granted…there’s a lot of other things taxing the filer where we have the VMWare data housed.
@Kevin Davis

]]> By: Kevin Davis http://ctistrategy.com/2009/07/20/deduplication-netapp/comment-page-1/#comment-1753 Kevin Davis Fri, 24 Jul 2009 19:19:20 +0000 http://ctistrategy.com/?p=289#comment-1753 I wonder if you could test this in a VMWare environment and see if the same results apply. We're taking a hit in performance (haven't dug too deeply yet to ascertain the cause) where we have VDI and LabManager housed on de-duped flexvols. Also, with all the increased OPS, did you happen to measure response time to see if it dropped off along with the increase in CPU load? I wonder if you could test this in a VMWare environment and see if the same results apply. We’re taking a hit in performance (haven’t dug too deeply yet to ascertain the cause) where we have VDI and LabManager housed on de-duped flexvols.

Also, with all the increased OPS, did you happen to measure response time to see if it dropped off along with the increase in CPU load?

]]> By: Jesse St. Laurent http://ctistrategy.com/2009/07/20/deduplication-netapp/comment-page-1/#comment-1750 Jesse St. Laurent Fri, 24 Jul 2009 17:58:23 +0000 http://ctistrategy.com/?p=289#comment-1750 Adam, This is absolutely the ideal workload for dedup. The data is completely redundant, it is 100% reads, and the deduplicated data set fits completely in cache. I was not trying to find the performance limit of the NetApp, but with the CPU running at 93% utilization, I expect there is not much headroom left. These results are for 4KB blocks. I have not run the tests, but I assume a smaller block size would bring the NFS ops up and the throughput down and a larger block size would bring the NFS ops down and the throughput up. - Jesse Adam,

This is absolutely the ideal workload for dedup. The data is completely redundant, it is 100% reads, and the deduplicated data set fits completely in cache. I was not trying to find the performance limit of the NetApp, but with the CPU running at 93% utilization, I expect there is not much headroom left. These results are for 4KB blocks. I have not run the tests, but I assume a smaller block size would bring the NFS ops up and the throughput down and a larger block size would bring the NFS ops down and the throughput up.

– Jesse

]]> By: Adam Leventhal http://ctistrategy.com/2009/07/20/deduplication-netapp/comment-page-1/#comment-1697 Adam Leventhal Tue, 21 Jul 2009 18:19:12 +0000 http://ctistrategy.com/?p=289#comment-1697 Very interesting. Is it fair to say that the workload chosen is nearly ideal for NetApp's dedup implementation and the results represent the effective performance limit? Very interesting. Is it fair to say that the workload chosen is nearly ideal for NetApp’s dedup implementation and the results represent the effective performance limit?

]]>