Comments on: iSCSI LUN is very slow/no longer visible from vSphere host

By: Sam

Sam — Wed, 17 Feb 2010 06:08:56 +0000

I have followed the advice in this post, but with no success…

I have been using Openfiler for 2 months now, 2 VMs (Windows 7 and Windows XP), both on one LUN.

Have had no problems until yesterday when my VMs just disappeared.

I reset the target as per the instructions here, but without success.

There is no heavy IO on these two VMs, so I don’t know what else to try.

Any help would be greatly appreciated.

By: It’s voting time.. « Virtualization, Windows, Infrastructure and all that “stuff” in-between

Mon, 04 Jan 2010 22:04:34 +0000

[...] “Lost” iSCSI LUN [...]

By: vinf

vinf — Mon, 23 Nov 2009 22:32:30 +0000

I can now repro this issue accuratley – my DC in a box has 10 x ESXi VM instances in it running on 2 physical hosts – if I kick off an HA situation by powering off one physical host the load of all 50% of the ESXi VMs rebooting seems to cause the disk reservation problems – each ESXi VM is mapped to use the same iSCSI LUN as the physical host, I’ve now distributed the hosts across 2 LUNs and all is well

By: Eric K. Miller

Eric K. Miller — Sun, 22 Nov 2009 05:19:38 +0000

We’ve seen the “locked LUN” issue on other SANs as well, not just OpenFiler. These are all Fiber Channel based SANs. Specifically, we’ve had more problems with a recent Infortrend unit we purchased (S16F-R1840). LUNs shared between as few as 2 hosts will experience LUN locks where the SCSI I/O Reservation never releases. This happens on both ESX 3.5 and ESX 4.0, and I’m assuming all previous versions. So, it’s not just an OpenFiler issue, but possibly a bug in ESX or a design flaw in the SCSI protocol or even the VMFS filesystem. I can understand the need for SCSI I/O Reservations, but everywhere I read, everyone runs into problems that ultimately result in downtime… not exactly a great thing to have to worry about.

Shrinking the LUN sizes and placing fewer VMs per LUN reduces the problem, but it’s sad that it’s truly a “gamble” whether the problem will happen or not.

Whether NFS is the answer or not, I’m not sure, but locks at the LUN level just seem like a horrible design.

Eric

By: Thomas

Thomas — Sun, 15 Nov 2009 18:04:12 +0000

I had locking problems with Openfiler and ESX in my lab. I noticed that in vsphere 4 under Configuration–> Storage Adapters –iSCSI adapter–> General Tab–> advanced button there are settings for iSCSI. Then I took a look on my Openfiler 2.3 box under Volumes –> iSCSI targets –> Target Configuration there were many of the same settings. By default there were a few that did not have the same values. I figured they should match and decided to change them on the openfiler box since I had multiple esx hosts in the lab. After making sure all matching settings had the same values….no more locking problems.

just a thought.

By: vinf

vinf — Fri, 13 Nov 2009 16:12:41 +0000

Hi – mm, I don’t know – my ML115′s have vmhba35 for iSCSI

By: HP ML115 G5 Autopsy (Motherboard Swap) « Virtualization, Windows, Infrastructure and all that “stuff” in-between

Tue, 10 Nov 2009 01:09:04 +0000

[...] in-between My ramblings on the “stuff” that holds it all together « iSCSI LUN is very slow/no longer visible from vSphere host HP ML115 G5 Autopsy (Motherboard Swap) November 10, 2009 Following on from my iSCSI [...]

By: mellerbeck

mellerbeck — Mon, 09 Nov 2009 18:19:53 +0000

quick question. You have vmhba34 for iscsi and mine is vmhba33. Do you know what determines this number? Or is it just based on what is in the system?

Thanks! -Michael

By: vinf

vinf — Tue, 03 Nov 2009 16:05:50 +0000

Funny you should say that… I’ve just purchased a 64 Gb SSD HD to try out in my OpenFiler box, blog post with benchmarks when I get time to try it out hoping to make use of linked clones to reduce disk space required by VMs and tier them with my normal bulk SATA storage in the OpenFiler box

By: deinoscloud

deinoscloud — Tue, 03 Nov 2009 15:05:22 +0000

Mate if you can afford it, go for SSD disks. I know they are expensive but if you do IOPS/$ math, you’re a big winner and no more SCSI reservation issues.

Cheers, Didier