FYI - Slow throughput issue resolved

[bdfry]

Just wanted to post a FYI concerning the resolution of an issue we were having with slow throughput on our backups.

Hardware:
Compaq Proliant 6400R - Windows 2000 (SP3)
Adaptec 29160 LVD SCSI Host Adapter connected to an Exabyte X-80 LVD w/ Mammoth-2 drives

Software:
ARCServe BrightStor v.9

Issue:
The average throughput on all backup jobs was averaging 60-80 MB/min

Resolution:
Changed the SCSI Host Adapter Driver to an earlier version. We are now averaging 600 MB/min for all jobs (some servers reported averages exceeding 1 GB/min)
We had previously installed 1GB fiber NIC's in the backup servers, but the throughput did not change.

Hopefully this info will help others that may be facing the same slow throughput issues that we were having...

Brantley

 

[Nightwatch]

Can you tell me what version of driver you went back to? Please include the date.

Thanks

paul

 

[bdfry]

Sorry about that... I should have included it in the original posting

We are using version: d4.2(4.10.4002) provided by adaptec
The data on the adpu160m.sys file is 2/21/2001

If you'd like me to send the driver to you, let me know..

Brantley

 

[Nightwatch]

Thanks,

I'm going to try and downgrade the driver of my 2940U2W and see what happens. How is the performance on you incrementals?

Paul

 

[bdfry]

Both the incrementals and fulls are averaging about 500 MB/min. We have a couple of servers at a remote location that bring the average down a bit on one of the jobs, but overall we've seen an increase 100 times what we used to see. Another piece of info that I should pass along is that after I changed the driver, I had to use the CA utility in the device management wizard to auto-discover the drives. Before doing this, I had very odd behavior when using BrightStor to inventory the library (a lot of unreadable media errors and slots being stuck in an update status). I also found that if you install the M2-monitor program on your backup server (assuming you are using an exabyte), you can use it to connect to each drive and perform tests. Before I downgraded the driver, I was unable to even see the drives when using the utility...

Brantley

 

[Nightwatch]

Glad to hear that. We've been struggling with poor through-put for a while. We adjusted TCP Window sizes and put all of our cards to Half Duplex. That seems to haveresolved a few issues. And to top it off we just had the Exabyte drive replaced this week. Can't say that this is been any fun. Incrementals are slow,; it tales 20 hr/s for 20 servers. I used to do it in 4 hrs with ArcServe 6.61. I'm going to downgrade the drivers as you did and see what gives. Currently it (the SCSI card) is using the Windows 2000 drivers. I don't know where you are but, if you hear a lot of swearing it's probably me

Paul

 

[bdfry]

I doubt I would be able to hear you swearing over my own...
This ARCServe/Exabyte combination has been quite an experience; many comp time hours have been accumulated because of it... We are looking at changing software to either Veritas of Commvault, backing up to disk (SAN or NAS), then archiving to LTO. But until all of that gets approved, I am stuck with what I've got... btw I'm in the deep south, Atlanta GA...

Brantley

 

[Nightwatch]

I've gone back to an older driver. Nothing worth noting seen. I'm going to try another one later on.

Paul
Montreal

 

[Nightwatch]

Well, I was able to resolve some issues on Friday. First I opened another call with CA. After explaining my problems concerning through-put they sent me a set of registry keys to modify. Before I did that I updated the Firmware on my SCSI card and reloaded the latest drivers from Adaptec. After a quick reboot I proceeded to modify the registry settings on the client agents. I only did 7 servers. The 7 are rather heavey servers averaging 40GB of data. Upon starting my Full Backup job I noticed that the throughput on one of the modified servers was twice what it used to be. This was also evident on several others. I think the problem was with the older Firmware on the SCSI card since the increased performance was also seen on servers whose registry was not touched. I did, however, have 2 servers that had extremely poor performance. I'm going to uninstall and reinstall the agents and see how it goes.

That's it for now.

Paul

 

[bdfry]

Excellent news !!!
If you don't mind, can you pass along the registry changes that CA recommended? I'd like to compare them to what I have on my servers.

thanks,


Brantley

 

[Nightwatch]

Sure thing, here's what they sent me:

Hello
This is with reference to your issue # 12694610, kindly check the following 1. Go to local area connections -> Choose configure -> Choose advance tab
and choose the Duplex -> check the drop box for full duplex. (Once the above settings is made the server will ask for reboot) 2. For more information please check the following website

http://support.ca.com/techbases/as2000/as2k-18140.html#soft

Try a test backup & check the through put, still if you have a problem then C
heck the following under regedt32.
Hkey-local-machine\software\computerassociates\Arcserveit\client agent parameters:

1. change the decimal value for Maxsendpacketsize to 8192
2. change the decimal value for RecDaemonbuffercount to 800
3. change the decimal value for Receivetimeout to 4800
4. change the decimal value for sendtimeout to 4800
5. change the decimal value for setwsrecdbuffersize to 64
6. change the decimal value for setwssendbuffersize to 64
7. change the decimal value for setwstcpnodelay to 8
# Select configuration and set the backup priority to high Stop and restart the NT client agent services . Please update the status of this issue through pcs@ca.com with the issue num ber in the subject field or give us a call at 866-422-2774. Have a good day. Please feel free to update your queries to:

http://esupport.ca.com/support_update/

Thank you


Let me know what your results are.

Paul