up2date, mirror repositories, and performance
Gene C.
czar at czarc.net
Wed Feb 18 20:13:13 UTC 2004
This is NOT a gripe!
I am posting this message in the hope that someone out there will have a
"better idea". I am not sure that there is a better idea but I can hope.
The current situation is better in some respects but it is still not enough.
When FC1 came out, up2date was configured to use a single server to download
files from. Even though Red Hat appeared to augment
download.fedora.redhat.com with heavy duty proxies (which it could only do
with the http protocol), it was not enough and download was slooow.
OK, FC2 and a new up2date with the capability to download from a mirror site
selected from a list of mirror sites. Better but still not enough.
There are a couple of problems with the mirror sites. First, they differ with
how much bandwidth they are connected to and how much loading is on a
particular server. Second, mirrors are not being updated "in a timely
fashion". That is, you may have an indication that updates are available but
when you run up2date, it connects to a mirror that has not been updated
recently. With the volume of updates occurring on a daily basis, it must be
difficult to keep the mirrors in sync.
Now, http is a connectionless protocol. You could download each file from a
different server. This has the potential of improving performance by
spreading the load more dynamically. Unfortunately, this requires a perfect
world where mirrors are always current.
Right now, my alternative it to manually identify a server with good
performance and a current set of packages. Then I edit sources to use that
server. I will likely switch to the approach I use with FC1 where I download
the stuff once to a local repository and then use up2date on the multiple
test systems I have to update from the local repository. All this is very
manual and I would like to have something which is more automagical.
The addition of the mirror list is a good capability but now suffers from
servers being out-of-sync.
Anyone have a "better idea"?
--
Gene
More information about the test
mailing list