Hyper-V Performance

Posted by bbrennan on 06-Dec-2016 16:25

Article Number 000010894 "DOES PROGRESS SUPPORT OR CERTIFY ANY OS VIRTUAL SERVER?"

Contains the follow note at the end of the article.

* Due to manner in which Hyper-V technology manages shared memory, Progress Software recommends that extensive performance testing be carried out, possibly with the vendor's engagement prior to production deployment, to verify that there is not a negative performance impact on OpenEdge applications.

Can anyone add some light on this.  We have just started working with a new customer running OE11.3 WG on Win2012 Server under Hyper-V.  Performance seems  poor for a small database.  I will be running through the database setup and parameters over the next day or so.  Before I get going too far, I wanted to check into the details on the article mentioned above.

I know about the NIC Card issues on VMWare.   Are there specific tuning details for Hyper-V?  I have looked here and in the kbase without a lot of luck.

Thanks,

Bob

OpenEdge RDBMS - Forum

All Replies

Posted by cjbrandt on 06-Dec-2016 20:30

It would be helpful if some db stats were available from a WG database on physical hardware to use as a comparison to the WG database on virtual.  

What is the record read rate from memory for phy vs virt.

What is the record read rate from disk - OS Reads for phy vs virt.

What is the write performance for the BI file or the database sync time during a checkpoint for phy vs virt.

Are the clients shared memory - or do they access data via the network ?

Trying to tune a db or hardware just going off a generic "things seem slow" makes it tough to know where to look.

Posted by bbrennan on 06-Dec-2016 20:54

Thanks for the reply. Your suggestions are the things I intend to hammer on in the coming days.  I am comfortable working though the needed tuning stuff.  It is puzzling to me that this small database with low activity seems to perform poorly compared to other similarly sized environments I have seen elsewhere.  It would be great to understand if there is some ‘feature’ of Hyper-V that will actively resist the normal tuning practices.
 
I am still interested in learning about the issues related to Hyper-V memory management that lead to the comment in the kbase article I referenced in my first post.
 
Bob
 

Posted by cjbrandt on 06-Dec-2016 21:02

maybe its a way of stating - Make sure your virtual environment isn't hosted on a NUMA server.  That is the only issue I have seen with OE interacting with the memory of an environment.

Posted by Rob Fitzpatrick on 06-Dec-2016 21:37

The configuration of the hypervisor's memory management settings can also affect guest workloads.  See this thread and the links in it for more info:

https://community.progress.com/community_groups/openedge_rdbms/f/18/p/9001/34493

The effects can be severe (like 15 minutes to proserve a db).  That was VMware/Linux but I have also seen some pretty horrific performance on badly-tuned Hyper-V, on good hardware.

I'd encourage you also to try to reproduce the performance issue(s) outside of OpenEdge and the associated application, e.g. with utilities or synthetic benchmarks to do disk reads and writes (sequential/random, buffered/unbuffered), network transfers, etc.  As Carey said, try to compare with a comparable reference system that has decent performance.

More than once I've seen admins refuse to help with a performance issue because "Progress is just slow", so the issue can't be theirs.  Reproducing outside of OpenEdge helps to stop that excuse before it starts.

This thread is closed