This post is more than 5 years old

1 Rookie

 • 

61 Posts

65027

March 4th, 2013 13:00

Does ODX accelerate Live Migrations on EqualLogic?

  • EqualLogic PS6100XS, 6.0.1 firmware
  • Server 2012 Datacenter, fully patched

Should I be seeing the benefits of ODX when I Live Migrate VMs? Because I'm not... :emotion-6:

1 Rookie

 • 

61 Posts

March 6th, 2013 22:00

Eureka! I finally figured it out. It's Veeam's fault - I posted on their forums with all the details:

forums.veeam.com/viewtopic.php

I'm pretty sure I was never actually seeing ODX transfers until I uninstalled Veeam. This begs the question of how I was seeing such incredible speeds, but I guess my SAN, combined with MPIO, can really fly even without ODX. It appears that the real benefit of ODX is how incredibly efficient it is at not using the network. With ODX enabled, my iSCSI traffic is virtually non-existent during these file transfers. The speeds are sustained around 200MB/s, which is definitely not shabby, but it's not the 700MB/s I was hoping for, either.

1 Rookie

 • 

61 Posts

March 4th, 2013 15:00

Thanks for the quick reply. Do you know for certain that 6.0.2 resolves any ODX issues? I didn't see anything in the release notes about it, so I assumed 6.0.1 would be fine.

1 Rookie

 • 

61 Posts

March 4th, 2013 15:00

I just upgraded to 6.0.2 and applied the hotfix, and still no dice. It's transferring very fast for a non-ODX transfer (about 125MB/s), but it's definitely not using ODX: the iSCSI network traffic lights up big time (which shouldn't happen with ODX) and ODX xfers should be way faster than 125MB/s.

1 Rookie

 • 

61 Posts

March 5th, 2013 07:00

We only have one array. See page 10 and 11 of this doc: http://i.dell.com/sites/doccontent/business/solutions/whitepapers/en/Documents/microsoft-offloaded-data-transfer-equallogic.pdf They clearly say there should be a 99% reduction in network traffic. Also, my copy speeds are about 125MB/s, and I should be seeing 600MB/s+

1 Rookie

 • 

61 Posts

March 5th, 2013 08:00

I don't see how that's particularly relevant. The paper clearly states that it's designed to offload the work from the network and onto the SAN, and that performance should be way beyond a typical transfer - nowhere does it say these benefits only apply to a 10GbE array. No offense, but it seems like you're arguing for arguments sake at this point. Unless you have something helpful to contribute, I'd appreciate it if you don't hijack this thread any further.

1 Rookie

 • 

61 Posts

March 5th, 2013 08:00

On page 8 of the Dell-provided whitepaper that I already linked to (i.dell.com/.../microsoft-offloaded-data-transfer-equallogic.pdf) it clearly shows a non-ODX transfer using iSCSI network traffic of 480MB/s send and ~700MB/s receive. On page 10, it shows an ODX-enabled transfer using 0kbps send/receive.

Why are you still debating this point? Do you think ODX doesn't actually help? Have you ever actually used it? What are you trying to prove here? If you're not interested in helping, please stop responding - I'm trying to actually figure out why ODX isn't working for me, not explain to you how it works.

1 Rookie

 • 

61 Posts

March 5th, 2013 13:00

This just keeps getting weirder. I just tested it again on the server where it had been working, and this time it started out at ODX speeds, and then dropped to non-ODX speeds. Note the green graph at the top drop from 600+MB/s down to 100MB/s:

I wish there was some sort of logging happening. I feel so blind!

1 Rookie

 • 

61 Posts

March 5th, 2013 13:00

Ok, I have an update, and apparently some humble pie to eat, too. I made several configuration changes to my 2 (clustered) servers, so I'm not sure which one made it work, but I am now getting ODX transfers on one of my servers, but not on the other. I'm pretty sure the change I made was to disable TCP/UDP Checksum Offload on the NICs (I also disabled it on the vNICs). A couple of weird things: 1) I made identical changes to both servers, but only one is working, and 2) (this is where the humble pie comes in), it does indeed show massive network utilization on both my iSCSI NICs during the transfer:

So, sorry, dwilliam - you were right. Apparently the documentation is wrong: there will be massive network utilization, at least as it's reported in the Task Manager. Here's what the other server looks like doing the same operation:

Furthermore, the whole reason I'm trying to get this working is to improve the speeds of Live Migrations, and can't get ODX to help Live Migrations at all yet - no matter which node the VM is on and/or which node owns the CSV, or any other variable I can think of trying yet.

1 Rookie

 • 

61 Posts

March 5th, 2013 14:00

I have a case opened, and given them piles of logs, including SANHQ. So far, they're stumped, too.

1 Rookie

 • 

61 Posts

March 6th, 2013 13:00

Even more weirdness. I discovered that there was a firmware and driver update available for the 331T (just came out very recently), so I installed those on the node (which I call node 2) where ODX was not working. After doing so, I had to completely reconfigure all the advanced properties of the NICs (I assume the firmware update wiped them out), and then I was getting AMAZING speeds:

However, the other node (Node 1) stopped getting ODX-enabled xfers at that same time. It uses the same NICs, so I decided to their firmware and driver. That got ODX working on that node 1 again, but caused node 2 to stop working. I kept fiddling with node 2, though. One thing that seems to "kickstart" the process is to initiate a move operation from Windows Explorer instead of a copy. After doing that, the next time I attempt a copy it seems to work, for whatever reason. But, you guessed it, then node 1 stopped working.

So, I've got a pretty replicatable problem right now: basically, if ODX is working on node 1, my next attempt on node 2 will fail, but then if I try again on node 2 after a little while it will work, and continue to work, until I try on node 1. The first try on node 1 will fail, but the next attempt will work, however that will cause node 2 to stop working. Classic whack-a-mole.

No Events found!

Top