I admit that I never thought of this case. Nevertheless, as far as I remember, the new caching algorithms in 6.5.x/7.x should be dynamic enough to cover up for this - maybe the reason your backup is slow is unrelated. Have you tried to collect a PerfStat output and analyze it yourself or with another person that is capable of reading its output?
Have you tried to dump to null to see if that is faster? There have been improvements in dump speeds across releases...
 
Eyal.
 

 
On 8/31/06, Carl Howell < chowell@uwf.edu> wrote:
Why not break the mirror change the minra setting and give it a go? You should know pretty quickly if it makes any difference. Our backup config is the same as yours, and we have a couple of volumes with millions of files that are also slow to backup.

- Carl
________________________________

From: owner-toasters@mathworks.com on behalf of Peter D. Gray
Sent: Thu 8/31/2006 6:53 PM
To: toasters@mathworks.com
Subject: Re: when to use minra ?



Thanks for the replies. Some background on this question.

I have 2 filers used for email storage (NFS, attached to 2
mirapoint mail stores). They have lots of small files.
I am mirroring these 2 filers to an R200 for backup, but
I also take tape copies via NDMP every 30 days or so. The volumes
contain about 350GB each of actual data, excluding snapshots,
and 5 million files.

My NDMP tape backups from the R200 are pretty slow. I average
about 2 MBytes per sec. Note that backups of other volumes
on te R200 work really well, so there is no problem
on the R200 or the network.

I saw the comment that turning off minra can have a big
impact on backup performance so I thought I would try it.
But discovered that the setting would have to be changed on the
mirror source which is not so good (maybe).

It would be nice if that setting could be changed on the
mirror target but not the source, but I can see why that not
possible.

So, I wondered if I could justify setting minra on on the
source volumes but would need some way to predict the impact.

It sounds like turing on minra is not usually a good idea, so I guess
I am stuck with slow NDMP backups.

Just out of interest, does anybody understand why minra has an
impact on backup speed. I assume the read ahead does not go
past end of file but that is the only explanation I can
think of as to why it would have an impact?

Regards,
pdg

--

See mail headers for contact information.






--
Yours,
Eyal.