Le mercredi 12 janvier 2011 à 08:05 +0100, Thierry Vignaud a écrit : > On 12 January 2011 04:29, Michael Scherer <m...@zarb.org> wrote: > > For example, distrib-coffee could blacklist you if you do this, if you > > are not alone on your network connexion. And when we deployed this > > measure to protect the server, the limit was 2 connexion per address, > > since this was taking too much ressources on the old server ( each http > > request taking 1 process and so memory ). > > switching to nginx would help that
According to the current mirror maintainer, the problem is not only on http server memory usage ( I do use nginx myself on my own servers and indeed, that would alleviate this part of the issue ). It is also that doing random access to file result in random seek on the raid array, and on a regular scsi disk ( not on ssd but ssd is not cheap for the moment ), this is more disrupting since the head of the disk is constantly moving to seek the block to fetch. Fetching continuous block from the disk is faster, since there is no delay. Fetching from random part of the disk is more problematic. And the same goes when done on multiple servers. It is lighter on the overall ressources to download on one mirror and stick to it than open 5 connexions and do 5 random seeks on 5 different disk. -- Michael Scherer