I have a similar setup and goal as Tim.
As another approach to the problem I've been pushing the sync's out to
the destinations rather than having the destinations pull the data. I
have 2G RAM on the source box (Solaris with a netapps disk) and I push
the data via rsync/ssh to destinations in parallel with a simple perl
script that forks a child per destination. (and loops over a few top
level dirs)
That doesn't really relate to Dan's question does it?? Oh well, I
guess my opinion is that if you can rsync using rsh/ssh (push or pull)
you will be much happier than if you have to use NFS.
eric
[EMAIL PROTECTED] wrote:
>
> One thing you can do to decrease your load is to ad the -W option. If you're
>reading it via nfs to do the checksumming, you have to read the entire file anyway,
>so you might as well just move the entire file, instead of wasting processor power,
>and
> reading the entire file twice (or more, actually).
> The single-processor bsd machine would max out at one transfer at a time, probably.
>using it as rsyncd, though, gives you the advantage of letting you use the "max
>connections" option, and having the individual machines retry until they succeed, thus
> controlling the load on the solaris machine. I am developing a similar solution for
>our system, where we have a single master copy of a set of tools, with identical
>copies all over the world that must be kept up to date.
>
> Tim Conway
> [EMAIL PROTECTED]
> 303.682.4917
> Philips Semiconductor - Colorado TC
> 1880 Industrial Circle
> Suite D
> Longmont, CO 80501
>
> [EMAIL PROTECTED]@[EMAIL PROTECTED] on 04/11/2001 03:08:28 PM
> Sent by: [EMAIL PROTECTED]
> To: [EMAIL PROTECTED]@SMTP
> cc:
> Subject: rsync across nfs
> Classification:
>
> I currently use rsync across an nfs mount.
> This nfs server is a sparc solaris machine
> mounting to approx 30 freebsd and 10 linux machines.
>
> When a typical rsync occurs to replicate data across all these
> machines....
> they all rsync
> /usr/local/bin/rsync -az --delete --force /home/cvs/website/config /website
>
> where /home/cvs is an nfs mount and /website is just the local drive.
> Problem is they all hit the solaris box at once driving it to load average
> as high as 75 for the 10-20 seconds that this occurs.
>
> My question is would i be better off taking a single processor freebsd
> machine....running a rsync server socket type deal...all getting 40
> machines or so to connect this way....or would that be worse.
> I like having all my machines update at once....just looking for an
> efficient way that a) i could even dedicate 1 box as just an rsync server
> and b) a single processor machine could actually handle that kind of load.
> My guess is i would prob have to stripe some drives together as IO may be
> a problem with that many webservers connecting at once? or will cpu be
> more a factor.
>
> I have considered changing rsync times around so that they don;t all
> connect to same server at once...this is done from crontab btw, but I like
> the way it currently is. Any suggestions would be much appreciated.
>
> --
> Dan
>
> +------------------------------------------------------+
> | BRAVENET WEB SERVICES |
> | [EMAIL PROTECTED] |
> | make installworld |
> | ln -s /var/qmail/bin/sendmail /usr/sbin/sendmail |
> | ln -s /var/qmail/bin/newaliases /usr/sbin/newaliases |
> +______________________________________________________+
