Posted by Joe Forani on October 03, 2014 at 14:03:32:
In Reply to: Re: Speed issues running Comet32 on Windows Server 2012 R2 posted by jim guerber on October 01, 2014 at 16:36:00:
The discussion continues ...
One big thing you are giving up by running on a desktop OS is volume shadow copies. They should be your first line of file-system level backups on a Windows system.
Desktop editions of windows are optimized to use multi-core CPUís differently than server editions. In particular the server editions use a CPU scheduler which offers better concurrency for multiple threads. Desktop editions are designed for maximum throughput on a single thread. For a dedicated comet host, a desktop OS is probably fine, but there are ways to tune server OS editions to achieve the same level of single-thread performance. However, you may be better off with a server OS on a comet anywhere host, as Iím not sure how Comet itself handles multiple sessions thread-wise. Jim could probably speak to that.
That is great. I have, for a long time, wondered why the full client could not perform as well as CometAnywhere. It has always been faster on the server and having CometAnywhere hosted on that server. Now as to your results on Windows7Öthat is sweet. What are the specs of the Win7 machine vs your server?
My only concern about running Comet off of a desktop OS vs a Server OS is that desktops need to be rebooted every once in a while. Servers are built to run 24/7. I reboot my servers a infrequently as possible. If you are not running in a 24/7 environment, then rebooting every once in a while is not an issue.
I tested Windows Server 2003, 2008 and 2012 in a VMWare environment. There was not a big difference in performance between any of those. I did not try running on Windows 7 or 8. Now you have made me curious. I think I may have a weekend project. :-/
Ok, a little bit of good news. We moved the CA host to the Windows 2012 server and TEST4 is giving us results of 2.13 secs on the host and 2.11 secs on CA clients - that's awesome. However, results on full Comet licenses sucks - 7.5 secs and higher. Tell me about your experience with Signature and this issue with CA performing better then full licenses. Right now it looks like a no brainer to trade in all the full licenses for CA licenses. What is the technical aspects for this performance difference? Is it just a case of CA clients running off the host?
Another thing we discovered - we switched the server to a Windows 7 on a core i7 and the same test blew away the Server 2012. TEST4 ran in .86 on the host and .96 on CA clients. Any idea why? We're thinking now that we could just make the Windows 7 box our server? Would changing the Server 2012 to Server 2008 get us the same results that we are seeing on the Windows 7 box?
Thanks again for your time, I really appreciate it.
p.s. Jim, I have copied you in here for your comments????
Yes, we used to run XAP and Comet Anywhere on a separate machine. After we upgraded, we moved XAP and CometAnywhere to the server. We have been working through some XAP issues, so temporarily it is on a separate box. I will be moving it back this weekend.
I would much rather support one box for Comet vs two or three!
Is your Comet server your CA host too?
Thanks for forwarding that to me. I have always been curious why the full client is so much slower than CometAnywhere. I never did get an answer to that question. We are in the process of migrating everyone to CometAnywhere so we can see the speed advantage.
Considering your Comet clients are 3.24 times slower than the server (7.5s vs 2.3s) is not too bad. Running that same test (1000.10) on our clients vs server resulted in a MUCH worse 6.47 times slower.
As you can see, the full client/network is where the problem is, not the server. Your 1000.10 times are only slightly slower than our 1000.10 times, but our server is significantly faster.
I am curious what the Test4 1000.10 times are on your old comet server?
We have reconfigured the server to RAID 10 and have not achieved any better performance. At this point it doesn't seem likely that we will try the bonded Ethernet cards or the SSD drives. Is there another solution? What about switching to Server 2008? The thing that is puzzling me (strictly from a non-technical point of view) is that the TEST4 program can perform so well (2.16 secs for 1000,10) on the server and not as well (by a long shot) on the network. Why shouldn't it? What is preventing Comet from pushing the data out to the network?
We are really stuck here. Any more insights would be most welcome.
Post a Followup
Each file can be a maximum of 1MB in length Uploaded files will be purged from the server on a regular basis.