We have a MD3000i connected to a iSCSI San for VM Ware, 4 ESX servers and 2 VC servers, connected to a 6905 cisco switch. Two controllers in the MD3000i and 2 NICs in each of the ESX servers and VC servers dedicated to iscsi. All of these are on one subnet for the iscsi. We are getting great reads but poor rights. About 1.4mb for rights and 3-4 for reads. A few questions
What is acceptable throughput for this iSCSI. What can be tweeked?
Thanks
Message Edited by quadrille on 07-30-2008 11:35 AM
I doubt you'll find an answer for acceptable throughput, but with VMware raid 10 is your best bet for performance, and obviously also a single virtual disk per disk group.
Check your Cisco documentation to see if there is any info about optimizing the switch for iSCSI.
Just a quick note, you write "All of these are on one subnet for the iscsi", thats a big "no no". Each port on the controller on the MD3000i should be on a separate subnet, so you will need two subnets. each nic on the client should be on different subnets. Like this:
Then you make sure that both nics logs on to both controllers, you can verify that from the MD3000i managment application. Done right you will have the same write and read speeds, and as an added bonus you will have complete loadbalancing, fail over and no single point of failure.
Read cache: Enabled Write cache: Enabled (currently suspended) Write cache without batteries: Disabled Write cache with mirroring: Enabled (currently suspended) Flush write cache after (in seconds): 10.00 Dynamic cache read prefetch: Enabled
Dell says this is not user configurable.
Those entries for the write cache saying "currently suspended" doesn't look right. Did support have any feedback on why it says that? It looks to me like it has suspended (effectively disabled) write cache, which could play a role in your very low write performance.
Write cache will be suspended if there is a battery error or if the battery is currently in a learn cycle (the latter only applies to the Gen 2 update which introduced Smart Battery support).
I am experiencing exactly the same problem. Here goes an excerpt from the profile data:
Battery status: Optimal Location: RAID Controller Module 0 Age: 27 days Days until replacement: Not applicable Last learn cycle: 14 January 2009 Next learn cycle: 15 April 2009 Weeks between learn cycles: 13 Part number: PN JY200 Serial number: SN 047118334386 Vendor: VN DELL Date of manufacture: 01 March 2008
Battery status: Battery learning Location: RAID Controller Module 1 Age: 27 days Days until replacement: Not applicable Last learn cycle: 14 January 2009 Next learn cycle: 15 April 2009 Weeks between learn cycles: 13 Part number: PN JY200 Serial number: SN 0471182J4767 Vendor: VN DELL Date of manufacture: 01 February 2008
Virtual Disk name: test-RAID1-100GB Virtual Disk status: Optimal Read cache: Enabled Write cache: Enabled (currently suspended) Write cache without batteries: Disabled Write cache with mirroring: Enabled (currently suspended) Flush write cache after (in seconds): 10.00 Dynamic cache read prefetch: Enabled
I guess there must be a problem with one of the batteries (although everything is marked as "optimal"): if the last learn cycle took place on January 14 and the next one is scheduled for April 15, why is it still in "learning" state?
Just a reminder that when you fail a controller you are supposed to do a "redistribute controller ownership / prefered path" on all the VD's handled by the MD3000i.....
It may have fixed it, but you may be forcing all I/O to the controller that was left online is all.
Thanks for the reminder! Indeed, I had forgotten to do it in the first instance, but luckily I noticed and corrected the mistake when I double-checked the configuration :)
Dev Mgr
6 Operator
•
9308 Posts
•
3 Points
1861
0
Posted July 30th, 2008 19:00
Check your Cisco documentation to see if there is any info about optimizing the switch for iSCSI.