On Mon, Aug 19, 2002 at 12:22:38AM +0100, Colin Morey wrote:
> Do we have a defined number of minimum connections a server should
> allow? or are we basing on an had hoc basis, based upon transit and
> availability as well?
That is an excellent question, and one I've yet to find a good answer
for. It is my opinion that we cannot afford to be too picky when
someone offers their services to us, but it also doesn't help to have
servers out there that can't perform. Unfortunatly, I've been too busy
with life to give this matter much thought.
The way I see it, we have three basic requirements for rsync mirrors.
1. Enough available bandwidth to serve a substantial amount of data.
2. Accurate and timely data.
3. Sufficient number of simultaneous logins to allow enough people to
actually use the mirror.
I would suggest that the mirror have at least a T1 of available
bandwidth. It doesn't have to dedicate the entire T1 to rsync
mirroring, but if all they have is ISDN, DSL, or cable, then they aren't
going to be pleased with rsync using all of their bandwidth. I would
suggest a minimum 128kb/s of available bandwidth for rsync bandwidth on
a mirror. This number would be better at 256kb/s, but 128kb/s should be
Some method needs to be developed for making sure that mirrors are updated
in a timely fashion, and have accurate copies of the master. There have
been several reports of out-of-date and bad data on mirrors. Right now,
we have people updating their mirror every 30 minutes on the 0 and 30.
This will quickly become inadequate if we continue to grow our number of
mirrors. Keeping an infrastructure like this healthy as it grows will
probably be the biggest task this admin team will face. I have lots of
ideas on this, but they can wait until we get some basic parameters down
for mirrors. Right now, it should be sufficient to develop an automated
system that checks for inaccurate and out of sync mirrors.
The third item seems to be the big point everyone is interested in, and
in my opinion the least of our worries. However, I'm tired of hearing
people whine about it so I've given it some thought. This ties into
item one since the more people you have using a server, the more
bandwidth you need. I think we should have ratios of bandwidth to
max concurrent users. Maybe 5 users for every 128kb/s of available
bandwidth. Basically, to determine these numbers we need to determine
the average user's rsync transfer, and come up with an average sustained
data rate for an rsync transfer.
Now, these are just my opinions. What do y'all think?