> Sort of, you need one helper 'slot' for each concurrent request. > > You can increase the number of 'slots' available by increasing the > number of children/helpers and the number of concurrency=N each can handle. > > at concurrency=1 you need 100 children for 100 requests, > at concurrency=2 you need 50 children for 100 requests > etc. > > This is mitigated further by caching the helper results for TTL=N time. > But worst-case is still 1 slot per request until the ACL result > mini-cache finds a duplicate. Ok, thx. I first thought squid had buffers (waiting queues) for helpers because of the "up to 1 pending requests queued" and "queue overload" messages. What do they mean? Also, what are the "negative lookups" of negative_ttl of external_acl_type? First I thought they were the ERR results, but apparently not. I tried some "stress" tests (ab -n 10000 -c 20 http://path/to/image.gif) and get aroung 770req/s. It seems low for a Xeon 3.40 Ghz with 3GB of RAM (cache_mem 2GB) and 200GB cache_dir on a RAID1 (with the system)... no? I tried to comment as much params as I could in the conf (removed siblings, store logs, etc) but it does not change anything... What's a normal number of reqs/s for such config? Also, while the url_rewrite logs lines would appear 10000 times, I only get like 14 external_acl logs... First 2 lines, then like 1 every seconds When I do x wgets, I get x external_acl logs. I have ttl=0, so it should not be a cache issue. Thx, JD
