Networker

Re: [Networker] Restore times excessive?

2003-01-07 19:24:31
Subject: Re: [Networker] Restore times excessive?
From: "Thomas, Calvin" <calvin.thomas AT NACALOGISTICS DOT COM>
To: NETWORKER AT LISTMAIL.TEMPLE DOT EDU
Date: Tue, 7 Jan 2003 16:21:42 -0800
We backup 45GB of data a night to networker. Time to backup: 3 hours.
We restore the 45GB of data. Time to restore: 3 hours.
I have had different speeds of backup / restore but mostly it appears to be
the disk subsystem, and the OS that play the greatest factor (besides the
network and that isn't a factor here).

Let me tell a story....
I was restoring our 45 GB of data to an Alpha server running raid 5.  One of
the disks was bad so it was running Raid 5 with no parity.  Our restores
took 3 hours.  We replaced the bad disk drive, and after the raid 5 had
repaired itself, I started running restores again to this system.  Our 3
hour restore now took 6.5 hours.  Since this was 3 months later, I didn't
immediately think of the Raid 5 system, but when I pulled one of the drives
out, the restore time went back to 3 hours.  We now run raid 0+1 on the
system and the restores are always 3 hours.

With Raid5, every byte of data written to disk has to do a Read of the
current disk data, add the new data to it, calculate parity then write back
to disk.  During a read it only has to read the data, calculate parity to
ensure there is no corruption, and then pass the data to the hard drive.  At
16MBS not many disk subsystems can keep up that pace continuously.

I am guessing your W2K system has a better hard drive subsystem since it's
backup is faster than the AIX system, or it isn't running Raid5.

Just for fun, try ftp'ing the same file between the two servers, and see
what transfer speed you get in both directions. What you really need to know
is the sustained read / write time of both disk subsystems you are running
these tests on.


Calvin Thomas
UNIX System Administrator
NACA Logistics






-----Original Message-----
From: Peter D. Gray [mailto:pdg AT UOW.EDU DOT AU]
Sent: Tuesday, January 07, 2003 2:13 PM
To: NETWORKER AT LISTMAIL.TEMPLE DOT EDU
Subject: Re: [Networker] Restore times excessive?


I think this is normal and nothing to do with networker.
Writes are a lot more expensive than reads. Try it
yourself with simple file read/write benchmarks.

It gets even worse with large numbers of small files.
File creation is very expensive. I allow a factor of
5 for restore over saves. Thats why we are moving to mirroring
for DR rather than tape restore. I extimate it would take a week just
to restore all our data.

Regards,
pdg

> I thought about parallelism. It's not a factor. I'm testing against a
single
> 10GB text file while no other operations are running on the NetWorker
> server.
>
> Dave
>
> -----Original Message-----
> From: Robert Legate [mailto:rlegate AT SHAW DOT CA]
> Sent: Monday, December 16, 2002 11:01 AM
> To: NETWORKER AT LISTMAIL.TEMPLE DOT EDU
> Subject: Re: [Networker] Restore times excessive?
>
>
> Could parallelism be a factor here? Are you staging or writing directly to
> tape? Multiplexing will occur when writing directly to tape and this will
> increase recovery time, since files are written in parallel and not
> sequentially.
>

--
Note: To sign off this list, send a "signoff networker" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=

--
Note: To sign off this list, send a "signoff networker" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=

<Prev in Thread] Current Thread [Next in Thread>