This post is more than 5 years old
1 Rookie
•
19 Posts
0
215554
May 25th, 2012 05:00
After some pointers on slow read speed over iscsi
Hello,
I have a ps4000, connected over iscsi to a windows 2008 r2 (hyper-v virtual) server,
I have some inconsistent performance issues with the read speeds in particular with small file sizes.
Copying large files to the san is fine, and copying small files to the san is also fine. However when trying to copy small files from either one iscsi drive to the another or from iscsi to local drives the performance drops dramatically.
It's much more noticable with smaller files, with the speeds dropping to as low as 10-15mb/s.
any ideas of where to start? I had looked at the disk cache settings, this is set to write-back which as far as i have read seems the better option for performance. The disk cache options are not available within windows, i assume as there are the settings on the san.
The same issue occurs when connecting the iscsi target to the host server (so it doesn't seem to be config issues with the virtual server).
There is a rise in cpu/nic activity when performing the copies but not to any noticable end.
If anyone has any thoughts they'd be much appreciated.
thanks
DELL-Joe S
7 Technologist
•
729 Posts
0
May 25th, 2012 07:00
Performance issues can be caused by several different factors (or even a combination of factors).
I would suggest starting with ensuring your SAN Network is setup (flow-control enabled, Jumbo (enabled if suppored by your switch and NIC), latest FW for the array, Host NIC’s and Switches, and updated drivers on the host. Start with this document: en.community.dell.com/.../2639.equallogic-configuration-guide.aspx (note that this document discusses general guidelines)
Task Offload may cause some issues in Hyper-V environment. Test by disabling them on the Guest Machine to see whether it helps:
In the NIC properties, in the advanced settings, disable all offload options (for example: Click the Large Send Offload Version 2 (IPv4) and change the value to Disabled. Do the same operation for other options, TCP Checksum Offload (IPv4), TCP Checksum Offload (Ipv6), Large Send Offload Version 2 (IPv6))
Also try disabling the global chimney as discussed here: support.microsoft.com/.../951037
Once you tried the global chimney and if performance is still poor, you may want to disable all the items discussed, to see if performance is better, than start to enable each to see which one may be the cause, For example:
netsh int tcp set global RSS=disabled
netsh interface tcp set global autotuninglevel=disabled
netsh int tcp set global chimney=disabled
netsh int ip set global taskoffload=disabled
If still having problem, please post your server (make/model, bios version) NIC/HBA (make/model, settings, driver and FW versions), your switch (make/model, settings, driver and FW versions), and the array (model and FW version).
-joe
JKHoffman-Seatr
1 Message
0
April 3rd, 2014 13:00
Just wanted to comment that these changes seemed to fix a performance problem I was having with a new Dell PowerEdge R320 running Windows 2012 connecting via iSCSI to an MD3220i. Some of those settings were already disabled (global chimney was for sure, but not positive about the others), but I ran all the commands, just in case. Rebooted the server over night and the next day the problems all seemed to have cleared up.
Thank you!
nickyg86
1 Message
0
August 15th, 2014 06:00
We was experiencing intermittent speed drops from 30-50MB/s to 0 bytes/s, but after performing those commands we no longer have speed drops.
Thanks Joe!
Dmitriy.Rezniko
2 Posts
1
February 6th, 2015 02:00
Update of hosting hyper-v server network drivers helped to resolve the same problem.