Unsolved
This post is more than 5 years old
7 Posts
0
31774
October 4th, 2011 03:00
EqualLogic PS6010XVS Poor Performance??
Hi Guys,
Im really struggling here with some performance issues on a VMware View environment which is using a PS6010XVS.
Im certain Im not over utilising the SAN for its quoted performance and have been working with a couple of the Dell EqalLogic guys in the US who have confirmed that the SAN itself seems to be operating and configured correctly....and I have also dealt with VMware who have said the config on ESX (full ESX 4.1u1) is sound also...
the ESX hosts are Dell M910 Blades with Broadcom 57711 10gbe nics (yes, I am running with the updated driver and also Delayed Acks are disabled too) - The Blade chassis has M8024's which then connect up to Cisco NEXUS 5548p's and then over to the Equallogic box.
My only thinking right now is to wonder if we have an error in our networking config??
So I wanted to get everyones thoughts on certain settings and what the appropriate best practices are...considering our position....
For instace...
10gbe - Flow Control..On or Off?
10gbe - Jumbo Frames....Yes or No?
To highlight the issue I am having....running ESXTOP with d is showing consistent DAVG in the high 20's and sometime hitting into the 30's and higher...KAVG is virtually zero!
If I dig deeper with ESXTOP, I can see that DAVG on read operations seems to be VERY high specifically.
Could this just be a product of flow control (which is currently off - hence the question, we had one guy tell us it wasnt needed on a 10g iSCSI network, hence why I want to find out the best practice for this?
Thanks
Paul Adam
Im really struggling here with some performance issues on a VMware View environment which is using a PS6010XVS.
Im certain Im not over utilising the SAN for its quoted performance and have been working with a couple of the Dell EqalLogic guys in the US who have confirmed that the SAN itself seems to be operating and configured correctly....and I have also dealt with VMware who have said the config on ESX (full ESX 4.1u1) is sound also...
the ESX hosts are Dell M910 Blades with Broadcom 57711 10gbe nics (yes, I am running with the updated driver and also Delayed Acks are disabled too) - The Blade chassis has M8024's which then connect up to Cisco NEXUS 5548p's and then over to the Equallogic box.
My only thinking right now is to wonder if we have an error in our networking config??
So I wanted to get everyones thoughts on certain settings and what the appropriate best practices are...considering our position....
For instace...
10gbe - Flow Control..On or Off?
10gbe - Jumbo Frames....Yes or No?
To highlight the issue I am having....running ESXTOP with d is showing consistent DAVG in the high 20's and sometime hitting into the 30's and higher...KAVG is virtually zero!
If I dig deeper with ESXTOP, I can see that DAVG on read operations seems to be VERY high specifically.
Could this just be a product of flow control (which is currently off - hence the question, we had one guy tell us it wasnt needed on a 10g iSCSI network, hence why I want to find out the best practice for this?
Thanks
Paul Adam
No Events found!
DELL-Joe S
7 Technologist
•
729 Posts
0
October 4th, 2011 07:00
The best practices for the EqualLogic array, is to have Flow control enabled on your switches (received desirable). Not sure if you reviewed these documents:
http://www.delltechcenter.com/page/EqualLogic+Configuration+Guide (Page 30)
Also, we have some documentation on VMware setup here:
http://www.delltechcenter.com/page/EqualLogic+Technical+Content
(search for VMware when on this page, you should see several documets)
You may also want take a look at the EqualLogic Multipathing Extension Module for VMware (MEM), to see if this is something that you might want to try as well. You can download this from the https://support.equallogic.com/secure/login.aspx (valid support contract and login name required).
Once logged in goto: Downloads> VMware Integration> EqualLogic Multipathing Extension Module for VMware® vSphere
Also check these as well:
Ensure you are NOT using the default vlan on any of the switches (create a unique vlan for iscsi traffic and ensure all devices in the iSCSI SAN are connected to it, i.e. VLAN2).
Jumbo frames could provide advantage, if both your switch and NICs support them (and are configured properly). However, note that some switches won't support Flow Control and jumbos, in which case Flow Control is far more important.
Ensure you have ESX “SOIC” (Storage IO Control) disabled.
Regards,
Joe
paul_adam123
7 Posts
0
October 4th, 2011 07:00
Thanks for getting back to me....looks like its the flow control thats going to be the killer as we have jumbo frames enabled, iSCSI is in its own dedicated VLAN, MEM is installed and SOIC is disabled!
Should flow control only be configured on edge ports (i.e. NIC switch ports and SAN controller switch ports?) or every switch port thats passing iSCSI traffic?
Bearing in mind that the traffic flow going something like this.....
vmnicxx
is connected to
port x on M8024 switch
then
port x on M8024 uplinks to
port x on NEXUS 5548
then
port x on NEXUS 5548 links to
SAN controller
There are two NEXUS switches which are linked with a vPC (virtual Port Channel) so should flow control be enabled on these ports too?
Cheers
Paul
paul_adam123
7 Posts
0
October 4th, 2011 08:00
Cheers
Paul
DELL-Joe S
7 Technologist
•
729 Posts
0
October 4th, 2011 09:00
Yes, flow control would be enabled on all your NICs/HBAs and other switch ports/connections passing SAN traffic. I'm fairly certain that on both the M8024 and the NEXUS 5548 you can configure both Flowcontrol and Jumbo at the same time (but check your switch doc’s or vendor first).
For the M8024 the minimum firmware is: v3.1.5.7
For the NEXUS 5548 the minimum firmware is: NX OS v4.2(1)N1(1)
On the array, ensure you are running the latest version of the FW:
If using v4.x - you should be at v4.3.8
If using v5.0.x - you should be at v5.0.8
If using v5.1.x – you should to be at v5.1.1-H1 (this is the ideal level of firmware to be at, but you can still use the other version listed if you are not ready to update to v5.1.x yet)
Regarding the interconnection between the M8024’s, and two NEXUS 5548's, how are they interconnected? You should have enough bandwidth for at least each interface used on the array(s). Example:
1 array member with 2 active eth ports, each at 10g’s, the ISL (or trunk) between each switch should be at least 20g’s (multiply this by the number of member/interfaces you actually have).
Regards,
Joe