On 8/24/2010 5:13 PM, Graham Frank wrote:
KB -- In general, CentOS mirrors have been fine enough in my experience. A simple test for something like this could be just a few samples of download rates periodically and then make note of it in the mirror status page.
Suppose an average response from a server of 80Kbps gets it a bad mark. 8Mbps or higher is good enough for most yum needs I would think. 32Mbps on up is excellent. Incorporating something similar into the status page and fastest-mirror could help get around issues such as this one.
We are not looking to come up with an index of how fast each mirror is. Rather, we are simply trying to identify problems with mirrors to preemptively handle issues and offer a baseline of a mirror's speed.
The options for this are numerous. Your thoughts?
Only problem I see with this is it would be dependent on what the mirror could pull from the other mirrors. Lets say the "checking" server is in the US, and the server being checked is in Australia. It might hand out 10Gb/s to local users, But it doesn't get used because its marked bad, All because the checking server is far away(network wise). Or lets say the "checking" server has some sort of bandwidth issue, Then they all get marked poor... Lots of things to consider on this one.
[1]: acceptable to the mirror admins. I dont think they would like it much if we were to pull a 600M iso every 6 hours as a speed test.
So why not pull a small 1MB test file? That's only a few megs a day, and less than a half gig a month. I dare say only testing once or twice daily would be necessary. You don't need much to see if any mirror is running at a crawl.
Hopefully my thoughts lead to something. Have a good evening!
--Graham
On Tue, Aug 24, 2010 at 5:59 AM, Karanbir Singhmail-lists@karan.org wrote:
I can fairly easily plumb in a scheduled regular tests that would check b/w rates from various places. However, the question then would be : what is a reasonable acceptable[1] performance test.
If we can come up with a test that is considered acceptable, we can then move to adding a metric from there into the mirrorlist generation to make sure that slower mirrors are used more infrequently.
The caveat here being that even through a machine might only be 10mbps, because its used more infrequently - it might give its 2 simultaneous users a better / faster experience than the 100mbps machine can to its 500 simultaneous users.
CentOS-mirror mailing list CentOS-mirror@centos.org http://lists.centos.org/mailman/listinfo/centos-mirror