On Thu, May 17, 2012 at 01:47:58PM +0200, Alfredo Finelli wrote: > Mark Hindley: > > OK, I accept that this is not working properly on faster systems (than I > > have). Actually the impact on response time and throughput is not as > > great as I had previously thought either. This is my proposed solution > > -- I would be grateful if you could test that it fixes it for you. I > > have implemented a new configuration option curl_throttle with a default > > of 1 (equivalent to 1ms). You could persuade me the default should be > > 10ms :) > > > > Mark > > Before I say anything let's see more timing tests: > > - apt-cacher (1.4.7): > > | | time (s) | speed (kB/s) | CPU % (4 min. avg) | > |-----------------+----------+--------------+--------------------| > | can_read(0.001) | 280 | 583 | 50.43 | > | can_read(0.01) | 249 | 656 | 2.35 | > |-----------------+----------+--------------+--------------------| > > - apt-cacher (your patch applied): > > | | time (s) | speed (kB/s) | CPU % (4 min. avg) | > |------------------+----------+--------------+--------------------| > | curl_throttle=1 | 327 | 500 | 50.41 | > | curl_throttle=10 | 308 | 531 | 2.31 | > |------------------+----------+--------------+--------------------| > > I think that those numbers speak for themselves: a value of 1ms is > increasing the global warming. So please, don't do it for me, do it for > the environment. :-)
Thanks. I will make the default 10. Is the patched version always that much slower or is that just related to what else was going on at the time? Mark -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org