Thank you very much for the feedback here guys! I believe we will drop the DVDs for now and see where that leaves us. Tru: Please change us to DVD=No.
We also have some bandwidth and connection limiting that is available under Plesk that we might be able to use. It does not look like Plesk will allow us to limit resource on a subdomain basis so we would likely need to use RayStedman.net instead of RayStedman.org for hosting the CentOS Mirror. This would also allow us to separate out our traffic stats from the main website a bit easier.
Mod_cband looks like it has a great deal of flexibility as well. I will try to read up on it over the next few days. Does anyone know if you can specify a subdomain as a limiting target?
We have several options to work with now. Thanks again! Greg
greg@raystedman.org wrote:
Thank you very much for the feedback here guys! I believe we will drop the DVDs for now and see where that leaves us. Tru: Please change us to DVD=No.
Why dont you look at your logs and see where the traffic is ? Perhaps an ISP has started using your mirror as a 'local asset' ? a polite email to them might help.
There hasent been any new iso release in the last few weeks, and I would be very surprised if the 5 fold increase in your b/w usage would be coming from generic dvd downloads. If you have limited b/w, dropping the DVD's would be a good idea anyway, I am just not sure if that is whats to blame in your case this time.
Mod_cband looks like it has a great deal of flexibility as well. I will try to read up on it over the next few days. Does anyone know if you can specify a subdomain as a limiting target?
yes you can, Also if you are going to consider dropping the b/w below 50mbps, please consider dropping the mirror completely.
Karanbir Singh wrote:
greg@raystedman.org wrote:
Thank you very much for the feedback here guys! I believe we will drop the DVDs for now and see where that leaves us. Tru: Please change us to DVD=No.
Why dont you look at your logs and see where the traffic is ? Perhaps an ISP has started using your mirror as a 'local asset' ? a polite email to them might help.
httpd access logs are inaccurate, unless you are using mod_logio or similar; they'll show a complete download amount for partial downloads, which causes AWstats and Webalizer to become *massively* inaccurate.
We had a situation with mirror.bytemark.co.uk a couple of months ago where someone was grabbing the first 1.5MB of a DVD ISO to test connectivity over their wireless mesh networks, over the course of 2-3 months it looked to Webalizer like they'd transferred >10TB.
Reality: bandwidth graphs of the switchport do not correlate, contacting them was fruitful, default access_log has serious flaws.
Hi Alex,
Alex Howells wrote:
httpd access logs are inaccurate, unless you are using mod_logio or similar; they'll show a complete download amount for partial downloads, which causes AWstats and Webalizer to become *massively* inaccurate.
Thats a good point. I've had a look through some of our ( mirror.centos.org ) aggregated logs ( we only do :80 to the users from there ), and I dont see any spike around the timeperiod Greg said he saw a spike. The OpenOffice update always causes a bit of a ripple, but that tends to settle down fairly quick. Also for new users who are installing from 5.1 media now, they would already have been seeing a previous OOo update, for them its just a case of getting a different rpm, almost exactly the same size.
Reality: bandwidth graphs of the switchport do not correlate, contacting them was fruitful, default access_log has serious flaws.
That does not make them useless or something to ignore, if you are suddenly getting a lot more traffic on the port, the first place to look would still be the logs.
-
Alex Howells -
greg＠raystedman.org -
Karanbir Singh