Hi,

Arunav Mandal wrote:

The director is running on xeon 2.8Ghz having 4GB of RAM on Hardware Raid0 and the storage demaon is on dual P4 3Ghz having 1GB ram and 1TB disk spool(xfs) also on Hardware Raid0. Most of clients are on P4 3gig machine with 1Gb ram running linux and windows.At peak there were atleast 40Jobs running at the same time. I will give a example of slowness below. As you can see the Rate is only at 568.2 KB/s before spooling it was around 8000KB/S. At the time of this client backup there were around 10 jobs running. As you can see this backup took around 7hrs. So my question is having latest hardware and enough memory why this is slow and what could be done to improve it.


You've got to consider other things as well.
First, the overall throughput - how long does it take all jour jobs to complete? Assuming that client speed is a limit, concurrent jobs should improve this. Second, network bandwith - If one client alone could saturate your backup server's connection, multiple concurrent jobs won't be faster. Your backup time window - sometimes it might be better to have clients spool data overnight and only wait for job completion (despooling) during work hours than having some clients waiting through the night and loaded by spooling data during daytime.

Arno

08-Sep 15:47  No prior Full backup Job record found.
08-Sep 15:47  No prior or suitable Full backup found. Doing FULL backup.
08-Sep 15:47 : Start Backup JobId 1012, Job=backup-abc.2005-09-08_15.47.00
07-Sep 15:48 Spooling data ...
07-Sep 21:41 Committing spooled data to Volume. Despooling 13,235,326,697 bytes ... 07-Sep 21:56 Sending spooled attrs to the Director. Despooling 122,253,435 bytes ...
08-Sep 22:12  Bacula 1.36.3 (22Apr05): 08-Sep-2005 22:12:50
 JobId:                  1012
 Job:                    backup-abc.2005-09-08_15.47.00
 Backup Level:           Full (upgraded from Incremental)
 Client:                 abc-fd
 FileSet:                "linux-default" 2005-09-08 00:20:01
 Pool:                   "daily"
 Storage:                "SDLT6K-1"
 Start time:             08-Sep-2005 15:47:02
 End time:               08-Sep-2005 22:12:50
 FD Files Written:       396,170
 SD Files Written:       396,170
 FD Bytes Written:       13,151,694,548
 SD Bytes Written:       13,208,137,447
 Rate:                   568.2 KB/s
 Software Compression:   None
 Volume name(s):         SDLT005
 Volume Session Id:      21
 Volume Session Time:    1126100490
 Last Volume Bytes:      154,989,306,412
 Non-fatal FD errors:    0
 SD Errors:              0
 FD termination status:  OK
 SD termination status:  OK
 Termination:            Backup OK




----- Original Message ----- From: "Arno Lehmann" <[EMAIL PROTECTED]>
To: "Arunav Mandal" <[EMAIL PROTECTED]>
Cc: <bacula-users@lists.sourceforge.net>
Sent: Thursday, September 08, 2005 9:09 PM
Subject: Re: [Bacula-users] Slow backups


Hi,

Arunav Mandal wrote:

I am running 1.36.3 on both director and storage deamon and diff. versions on 150 clients. Recently I started spooling to disk first then to tape with around 100 concurrent connections for director and storage deamons. Now the backups seems to take time Rate seems to be around 700KB/s before spooling and concurrent connections it was around 8000KB/s. I know too many clients are backing at the same time is that the reason?


I can't follow you here, but yes, having too many clients working concurrently can reduce the overall performance. You've got to find your personal optimal number of concurrent jobs depending on overall network throughput, server power, storage and spool speed - I don't think that it's possible to tell you what the perfect setup is for you.

Also when all the clients start backing up at the same time I can't log onto the console since there is no free connections left is there any way I can atleast reserve one connection for the console?


Two possible reasons:
Not enough connections allowed - change the numer in the config.
Server load too high - wait, or, better, have less clients run simultaneously. Remember that each job means a fork and a process, takes memory, cpu time, and loads the catalog database. That's all stuff you need to work on the console, too.

Thx,

Arunav.



Arno


--
IT-Service Lehmann                    [EMAIL PROTECTED]
Arno Lehmann                  http://www.its-lehmann.de




--
IT-Service Lehmann                    [EMAIL PROTECTED]
Arno Lehmann                  http://www.its-lehmann.de


-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf
_______________________________________________
Bacula-users mailing list
Bacula-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bacula-users

Reply via email to