VMTN
Blogs
Twitter
Facebook
YoutubeExchange Performs Well Using Fibre Channel, iSCSI, and NFS on vSphere
In a new whitepaper, a large installation of 16,000 Exchange users was configured across eight virtual machines (VMs) on a single VMware vSphere 4 server. The storage used for the test was a NetApp FAS6030 array that supported Fibre Channel, iSCSI, and NFS storage protocols. This allowed for a fair comparison of these three storage protocols on the same hardware. The test results show that each protocol achieved great performance with Fibre Channel leading the way, and with iSCSI and NFS following closely behind.
Similar tests have been done to compare Fibre Channel, iSCSI, and NFS on ESX in the past. These tests used IOmeter to measure the storage performance. In this new round of tests, Exchange Load Generator was used as the test tool to simulate an 8-hour work day.
The results show that Fibre Channel provided the best performance with the lowest CPU utilization. Additionally, iSCSI and NFS were relatively close in performance. The two graphs below summarize the test results showing the sendmail average latency as reported by LoadGen and the overall CPU utilization of the the ESX server.
The complete whitepaper has all of the configuration details and additional test results.
I'm curious to know if the iSCSI and NFS options were tested with a single 1GbE connection or 10GbE.
Posted by: Rick Scherer | July 23, 2009 at 01:12 PM
The answer is that a single 1GbE connection was used for the iSCSI and NFS tests.
In the spirit of a friendly blogger I would like to say that this info is in the whitepaper along with info regarding a test done with 4 x 1GbE for iSCSI.
Thanks for the question.
Posted by: Todd Muirhead | July 23, 2009 at 02:25 PM
It doesn't seem fair to compare these protocols when using a NetApp box that virtualizes FC and iSCSI LUNs as files within the WAFL file system. Seems like you're throwing away some of the inherent low-latency, high-bandwidth benefits of FC with the additional overhead of a file system and the possibility of file (LUN) fragmentation.
Posted by: Dave Boone | July 27, 2009 at 11:11 AM
Dave, a little FUD coming from EMC? Every modern storage system uses virtualized data objects. Whether you call them "Meta LUNs", or files, the real test is the effectiveness of the design to provide the performance, efficiency, and ease of use required by the customer. NetApp delivers on all three.
Posted by: Jason Blosil | July 30, 2009 at 06:43 AM
Is there somewhere same kind of storage protocol test using EMC Clariion/Celerra combination? Is it possible that NetApp WAFL will slow down FC and iSCSI?
Posted by: Mika Ollikainen | August 12, 2009 at 12:18 AM
I think - for NetApp it is normal that FC option shows such bad results. Remmeber - FC device on NetApp is a file. That's one more abstraction level. They must test other vendors mid-range platforms EMC, HP, HDS. The results are for NetApp not for vShpere...
Posted by: Pavel Filin | September 09, 2009 at 04:48 AM
@Dave Boone - I realize my follow up is 'somewhat' dated - but I believe your concerns about NetApp virtual LUNs is unfounded.
In fact, in this test the performance of every storage protocol out performed the same set of tests completed on EMC FC.
Now the tests ran on EMC were with VI3 (as opposed to vSphere). While vSphere provides increased IO performance it does not equate to the delta between the tests nor does it explain why 'real' EMC FC LUNs underperform virtual NetApp LUNs and NetApp NFS.
Posted by: Vaughn Stewart | February 10, 2010 at 04:15 PM
Any chance the type of NIC used is available ?
To see if it does TOE offloading for iSCSI ?
Posted by: scott owens | March 05, 2010 at 09:28 AM
The NICs used were Intel PRO 1000 NICs based on the 82571EB controller. It does not do TOE offloading for iSCSI.
Thanks - Todd
Posted by: Todd Muirhead | March 05, 2010 at 08:38 PM
On page 47 from Kens Virtual Reality - http://kensvirtualreality.files.wordpress.com/2009/12/the-great-vswitch-debate-combined.pdf
he writes that you can get greater iSCSI performance by moving the iSCSI initiators into the VMs themselves instead of letting the ESX host do the work.
He does point out that this initially is greater admin work but the load balancing & performance are increased - so is CPU but that does not seem to be an issue.
I wonder what kind of improvement that would have shown.
Posted by: Lonnie Cingular | March 06, 2010 at 06:36 AM
Lonnie, thats pointless and avoids half of the benefit of storage virtualisation and therefore virtualisation itself. But if performance is more important than ease of use, management or cost of ownership, then cool.
Posted by: Spud | March 11, 2010 at 03:35 AM
You can have better performance if you will use FCoE data transfer protocol. Running at 10Gb Ethernet and Fibre Channel at 8Gb will improve you infrastructure. We tested in on HP WM.
You can see FCoE detail at http://fcoe.ru
Posted by: Alex | March 22, 2010 at 06:07 AM
I just finished reading . I liked it. I’m glad I saw this post on ; and that I am adding it as my favourite. This blog on has helped me in getting additional information about & also helped me see big picture context, which is valuable. It gave me a feel, and a long term view of
Posted by: Bulk SMS | May 22, 2010 at 01:18 AM