Saturating Cisco 2960s stack with iSCSI traffic; move to 3560g? 4948?

hotcrandel

Gawd
Joined
Feb 26, 2010
Messages
781
I imagine for some of you this would be a fairly basic question;

I deployed a ESX cluster a few years ago on a 2960s stack. HP G6 dual quad core, each server with 8 1gb NIC's. HP P4300 storage

It was relatively low traffic and seemed to work well from what I can tell.

I've since added a new Tegile Zebi SAN product with SSD caching and whatnot, as well as added Veeam backups.

My 2960's seem to be dropping a lot of traffic, in particular from the Zebi. I've read that this is because that iSCSI traffic, especially a faster san can burst traffic too fast for the switch.

I've been looking up this and other people seem to have the same dilemma; latency and packet drops from 2960's, and moving into a better switch or two.

Would a 3560g be better? Everyone seems to say that 4948's are better than those as well; Looking at the prices on ebay and things they don't seem to be too expensive. ($1000-->$2500)

Any other advice is welcome as well.
 
I run my Equalogics on a set of those. They are set to be replaced this year because of this issue.
 
In my last position I was involved with a rather large School District had 6 EQL nodes/bricks on a pair of 2960's. Needless too say that they had extreme performance issues and it was just that the buffers were too small.

EqualLogic would not continue support until we replaced them with supported switches. In my current job, I still see it, people are still architecting EQL storage with switches that are not on the Tier I supported list. I have since given up driving that home...yet again, here is another example of that.
 
In my last position I was involved with a rather large School District had 6 EQL nodes/bricks on a pair of 2960's. Needless too say that they had extreme performance issues and it was just that the buffers were too small.

EqualLogic would not continue support until we replaced them with supported switches. In my current job, I still see it, people are still architecting EQL storage with switches that are not on the Tier I supported list. I have since given up driving that home...yet again, here is another example of that.

Yeah we had to use what we had when we implemented them, but we are in process of grabbing a few 4948E's.
 
FYI the 4948 hasn't dropped any packets at all since installation. Seems to have solved the issue.
 
Thanks for posting the resolution. Not many post back after getting advice on whether it was good or not.
 
I've run into similar, but only doing some things that most would consider "absurdity" :)

(Violin 6616, small frames, LOTS of scsi commands, the Nexus 5k finally told me to go pound sand).
 
I've run into similar, but only doing some things that most would consider "absurdity" :)

(Violin 6616, small frames, LOTS of scsi commands, the Nexus 5k finally told me to go pound sand).

lol

Hey, these things are so fucking expensive, they should be able to do all of that AND compute the meaning of life.
 
Back
Top