it might worth trying to change few bits in the source code and implement this feature. I thought about adding 'tos' field to squid reply_header structure and read this value from source. However , squid doesn't deal with packets, it deals with HTTP requests/replies. in our case ,how do you guarantee that all packets through current connection hold the same TOS ? Is it possible for squid the inspect all packets ? I don't think so. On Sat, Apr 23, 2011 at 4:24 PM, jiluspo <jiluspo@xxxxxxxxxxxx> wrote: >> On Sat, 2011-04-23 at 20:36 +0800, jiluspo wrote: >>> >>> remote servers I mean http web servers TOS. >>> I already know about peers in fact current squid(as of 04/24/11) TOS are >>> not >>> being marked peer(digest or icp) hit when local miss. >>> http://bugs.squid-cache.org/show_bug.cgi?id=3202 >>> >>> AFAIK squid 2 series TOS always marked zero from remote servers. >>> according to source code initial tos=0; >>> >>> there are some patches called preserve tos miss but kernel(linux) needs >>> to >>> be patched. >>> >>> does kernel really need to patch in order to pass the TOS value from >>> kernel >>> to squid? >>> >> >> Yes, I'm afraid it does, due to the way the networking stack works. >> >> If you want *similar* functionality *without* patching the kernel, then >> you can use the "qos_flows mark" feature, which uses the netfilter mark >> value rather than the TOS value. However, marks do not apply remotely, >> so this will only work to retain marks on the local machine. > > therefore squid 3.2 still cant preserve TOS value from remote server to > clients. > hmn. what about the zph that requires kernel patch? would it work with > remote servers? > > lastly, what about its performance degradation(req/sec and service time) if > we add this feature. > has anyone run some benchmarks if with/without mark. > > -- > This message has been scanned for viruses and > dangerous content by MailScanner, and is > believed to be clean. > >
