Fixed path faster than round robin

[paulow1978]

just a quick update:

I have not tried the direct connection yet as I have some vm's in use on the equallogic.

But....

I have stopped starwind and installed the microsoft iscsi target and tested that out. Still no better so this is def not a software san issue.

I updated the onboard broadcom nics and they have jumbo frames now. So set up them to be used istead of intel card. Again no luck.

this is just bizarre. The next things to rule out are the switches......

[paulow1978]

Hi,

Saw your message on my post on the vmware forums.

I am still investigating but essentially I seem to be getting somewhere now.

I have built a new ESXi5 server, set up round robin and linked it to the starwind SAN. I have moved this set up off my Avaya switches and done some performance testing using a netgear GSM724t with no jumbo frames enabled on the vswitch and iscsi set up or switch.

Performance has rocketed up to 200+mbps.

That leaves me with two reasons this is happening..

either the iscsi mpio on my first ESXi server is not working as expected or the avaya/nortel switches are not behaving correctly.

Next step:

I am going to move over the physical cables that are plugged in to the netgear back on to the nortel/avaya switches and then re run the performance tests and see what the results are. This will rule out it being an ESXi iSCSI problem with the first ESXi server and point to the switches.

thanks,

Paul

[paulow1978]

Right,

issue is to do with our Avaya/nortel switches. 10k's worth of switches in a stack and a netgear out performs it!

I am looking in to this issue now.

thanks,

Paul

[anton (staff)]

Like I've predicted Hope you'll manage to solve the issue. Please keep us updated. Thank you!

Right,

issue is to do with our Avaya/nortel switches. 10k's worth of switches in a stack and a netgear out performs it!

I am looking in to this issue now.

thanks,

Paul

[paulow1978]

Ok guys,

well I finally got to the bottom of it. I will post the resolution in case anyone else out there has issues with nortel/avaya switches...

The nortel/avaya switches buffers are linked to each group of 12 ports. By default each port is only allowed to use up to 33% of the buffer. This has to be changed to Maximum. Also flowcontrol has to be set to auto and the QOS queue depth to 1 if using any other switch than a 5600 series (otherwsie it's 2). SPanning tree to be disabled on all iscsi ports.

The commands to enter on the nortel are

(config)# qos agent buffer maximum
(config)# qos agent queue-set 1
(config)# interface fastEthernet ALL
(config-if)# flowcontrol auto
(config-if)# exit

then shutdown command to reset the switch (needs a reboot to take effect)

there is an avaya doc on all this but if you search google/forums etc its very difficult to find (like all avaya/nortel stuff)

http://downloads.avaya.com/css/P8/documents/100123892

I can now tell you that my iometer mpio is up to 218mbps on round robin using starwind

many thanks,

Paul

[Anatoly (staff)]

Wow! That was tough

OK, it was really our pleasure to assist you in this investigation.

Feel free to contact us if you`ll have some more questions!

Thank you for your cooperation!