VMware Cloud Community
YannickN
Contributor
Contributor

After upgrade to esx 4.1 iscsi traffic brandwidth is slower

Hello,

We have for many of our customers, made a similar architecture VMware / Netapp, composed of vSphere 4 and FAS2020A.

We used the iSCSI protocol and ethernet cards with 1 gb / sec. Each storage controller has 2 Ethernet ports of 1 GB / sec. and each ESX has two cards 1 gb / sec didicated for storage network connected to a different physical Switch. So we have two physical path to each LUNs.

We have set the PSP Round Robin on our LUNs.

During our test with vSphere4.0 we had a rate of 120 MB / sec. Approximately 60 MB / sec on each ethernet link.

Since the passage in version 4.1 the rate is now of 60 MB / sec.

When

we look at the graphic use of our switch, we can see that our two physical

links are used to 25% in contrast to 50% in version 4.0.

We don't use vdistribuded swith, so the specific QOS of vSphere 4.1 does not be used.

After extensive research we still have no solution.

Is this a known issue ?

Thank you for your help.

0 Kudos
3 Replies
ExpletiveDelete
Enthusiast
Enthusiast

possible path thrashing? try setting your paths to 'fixed'.

0 Kudos
YannickN
Contributor
Contributor

So i have made a new test with iometer with a file of 600 MB, in fixed it use 50 % of one link. And in round robin 25% of the two link.

Why did esx not use 50 % of each link in round robin ? is my 600 MB file to small to use more than 50 % of one link ?

It's the same problem when I deploy a big template (40 go) in round robin it use 25% of two links, not more...

0 Kudos
YannickN
Contributor
Contributor

With a fas 2040 which has 4 ethernet port when we deploy a vm with a disk of 40 GB it also use 50 % of the two links from the esx server and 25 % of the four links of the storage processor.

So it seems to be the same limitation. Is there a limitation of brandwidth on the esx site for the iscsi initiator ? or per iscsci session ?

Thank you in advance.

0 Kudos