> Weibin Yao wrote: > > I'am using squid-2.7. I has checked the configure reference and found > > nothing about sendfile(). Why squid does not support sendfile()? > > especially the HIT request. On 26.01.10 21:26, Amos Jeffries wrote: > 1) Blocking call. Squid needs to support more than one client request > simutaneously. is it blocking anywhere? > 2) speed. sendfile is limited linearly by disk IO speeds, blocking the > entire time. does it matter for content fetched from disk? I think that sendfile is for this cases the most effective option (e.g. from disk direct to network card memory). I understand it can be an issue in 3.x where squid wants to implement own caching, but wonder if sendfile couldn't help here as you indicate. > 3) HTTP protocol. The current design of Squid stores the headers and > data together. They cannot be altered correctly according to protocol > requirements during a sendfile() call. you can read, process and write headers and THEN call sendfile for the rest of content. The problem is with chunking which it not supported on client connections yet, iirc. > 4) collapsed forwarding. multiple clients may be receiving the same > identical object from Squid simultaneously, or even different parts of > the same object. should not be a problem with sendfile, should it? > 4) object location. not all HIT objects are from files. some may be in > memory, or a range of something partially received by another client. 5) I think ;-) Yes sendfile is only applicable on content fetched from the disk. Apparently nobody implemented sendfile in squid yet and apparently nobody will do it, but I wonder if all those reasons are really that problematic... -- Matus UHLAR - fantomas, uhlar@xxxxxxxxxxx ; http://www.fantomas.sk/ Warning: I wish NOT to receive e-mail advertising to this address. Varovanie: na tuto adresu chcem NEDOSTAVAT akukolvek reklamnu postu. Windows found: (R)emove, (E)rase, (D)elete
