Performance of GP 2013 2.5 Times Slower on Virtual Servers than Physical

(0) Share

Report

Posted on by Community Member

I have found something that is perplexing me. In particular with Smartlist performance, I have see a drastic drop in performance when moving from a five year old Dual Processor Dual Core Dell PE 2950 with only 8GB of RAM and a RAID 1 and RAID 5 disk layout to a Virtual Environment with 16 virtual cores, 64GB's of RAM and utilizing a Compellent SSD disk backplane (datacenter grade). The old server produces Smartlist results 2.5 times faster than the new WMWare Virtual Server. I have seen this happen in more than one environment in two different companies who moved from physical to virtual when they upgraded to GP 2013. You see the same results with GP 2010 R2 before the upgrade to 2013.

Has anyone run into this drop in performance? Note, memory, processors and disk IO's barely get pegged on the new server when the Smartlist is producing results. Also, on the old server the results run at a consistent speed but on the new server the speed fluctuates.

*This post is locked for comments

I have the same question (0)

All responses (22)

Answers (1)

Jonathan Fear on at

Like (0)

Report

Have you checked out the Performance White Paper yet?

mbs.microsoft.com/.../MDGP2010_WhitePaper_Performance

Was this reply helpful? Yes No
Community Member on at

Like (0)

Report

LOL. Of course. After I hit submit on the post I regretted not putting that in the facts. I knew that would be the first response. :)

Was this reply helpful? Yes No
Community Member on at

Like (0)

Report

Tried various hyper/multi threading settings to no avail. There is absolutely no contention whatsoever on any hardware resources using all measures in the doc as well as some that we know of above and beyond those.

Was this reply helpful? Yes No
Josh Thompson 1,115 on at

Like (0)

Report

how are ping times from the client to the server?

can you perfmon the network stats on the sql server and see if there appears to be a bottleneck?

How about looking at the execution plans between the two servers before and after the upgrade?

Have you done typical db maintenance? ie. update stats, reorg/rebuild indexes as appropriate?

Did the SQL server itself change at all?

Was this reply helpful? Yes No
Community Member on at

Like (0)

Report

This is a situation where we have migrated from a Physical server to a Virtual server, so yes SQL has changed I guess you could say. We went from SQL 2008R2 to 2012, but we have also tried this on 2008R2 with the same results of degradation. It doesn't matter if you run Smartlist from the server itself or a workstation so network connectivity is not an issue. Full DB maintenance has been done with no impact. We have traced disk I/O and it barely makes any noise on the disks. Memory is hardly impacted and processor barely shows any impact.

This is obviously an issue with Virtualization not working well with Smartlist technology. Again, we have migrated from Physical to Virtual at two sites and the same results. We set up VMware at one client and at another their certified VMware specialist set it up. Two approaches, same technology, and same results = degradation of Smartlist performance by 2.5 times.

Was this reply helpful? Yes No
Steve Kubis on at

Like (0)

Report

Because you are on a virtual machine, there's one more issue you should consider. This information comes from the Dynamics GP Web Client Installation and Administration Guide:

Another situation that can occur with virtual machines involves Network Interface Card (NIC) settings that are not fully compatible with the operating system settings. Specifically, NIC settings may be set to use “offload” optimizations that actually slow down network performance for the virtual machine. To turn off these optimizations, do the following on each virtual machine.

1. Open the Network and Sharing Center.

2. Click the link for Connections to view the information about the local network connection.

3. Click Properties.

4. Click Configure.

5. Click the Advanced tab.

6. Disable all of the settings that have “offload” in their name.

7. Click OK to save the changes. This will reset the NIC for the virtual machine.

I have seen cases where GP performance is noticeably worse on a virtual machine, but when you make the changes described in this procedure, performance returns to normal on the VM.

Note that you're making these changes in the operating system of the VM, not on the machine that is hosting the VMs.

Steve

Was this reply helpful? Yes No
Suggested answer

Josh Thompson 1,115 on at

Like (0)

Report

To further what Steve said, you may be able to validate some of this by also looking at the waits on SQL Server. if you are running a lot of smart connect and seeing lag, look to see how your ASYNC_NETWORK_IO waits are. Those typically indicate that SQL has finished processing and it is watiting for the application/network to deliver and consume the results.

Was this reply helpful? Yes No
Community Member on at

Like (0)

Report

Are you saying that even though I am not running GP over the network (running Smartlist on the client on the SQL server itself) that network could still be an issue? Are you saying that SQL loops back over the NIC even though the SQL Server is on the same machine as the GP client?

Was this reply helpful? Yes No
Community Member on at

Like (0)

Report

Steve,

I tried your suggestion and it had no impact.

Josh,

The ASYNC_Network_IO numbers are similar between servers.

I also verified all Performance measures in the Optimizing Performance White Paper and there are no bottlenecks shown in any metrics.

Was this reply helpful? Yes No
Suggested answer

Richard Whaley 25,195 on at

Like (0)

Report

In VMware, check the amount of RAM assigned to the virtual machine. The default is 2gb but often more is needed.

Was this reply helpful? Yes No