On Thu, 6 Dec 2007, Tharindu Rukshan Bamunuarachchi wrote: > Dear Stephen, > > Thankx for the quick reply.... > > On Dec 6, 2007 10:40 AM, Stephen Hemminger > <shemminger@xxxxxxxxxxxxxxxxxxxx> wrote: > >>On Wed, 5 Dec 2007 23:58:59 +0530 "Tharindu Rukshan Bamunuarachchi" > <tharindunix@xxxxxxxxx> wrote: > >> > >> Dear All, > >> > >> How can I maintain constant/minimum socket receive buffer. > >> > >> I have used setsockopt() with SO_RCVBUF value for 256bytes > >> > >> > >> > >Why? Is your application limited? or is the network so busted it can't > >take outstanding data? > > - We are trying to throttle 'sender application' by not reading from TCP > stack (from receiver side) after reading certain number of messages. > - Here we wanted to limit number of data buffered in the receiver side when > receiver application stops reading from TCP stack. > > >Setting RCVBUF less than MTU size is likely to be ignored, otherwise > >the sender on the other > >size would be do silly window avoidance. > > - we set SO_RCVBUF value to 512 bytes before listen, then TCP negotiated 768 > bytes while establishing a connection. We stopped reading from TCP stack > after some time and we observed that TCP advertising window size reduced to > 0. Subsequently receiver advertised higher values such as 1152, 2304, 1460. > - Does this mean that receive buffer size has grown beyond initially set > value (512 bytes)? > > 11:48:46.887136 IP v20z-10.28902 > 172.25.41.159.2963: S > 1720573913:1720573913(0) ack 253836197 win 768 <mss 1460,nop,nop,sackOK> > 11:48:46.894547 IP v20z-10.28902 > 172.25.41.159.2963: . ack 113 win 656 > 11:48:46.897735 IP v20z-10.28902 > 172.25.41.159.2963: P 1:92(91) ack 113 > win 656 > > 11:48:47.295885 IP v20z-10.28902 > 172.25.41.159.2963: . 144498:145958(1460) > ack 3321 win 11 > 11:48:47.328583 IP v20z-10.28902 > 172.25.41.159.2963: . ack 3321 win 1152 > 11:48:47.329513 IP v20z-10.28902 > 172.25.41.159.2963: . ack 4473 win 0 > 11:48:47.331274 IP v20z-10.28902 > 172.25.41.159.2963: . ack 4473 win 1152 > 11:48:47.332186 IP v20z-10.28902 > 172.25.41.159.2963: . ack 5625 win 1152 > 11:48:47.333379 IP v20z-10.28902 > 172.25.41.159.2963: . ack 6777 win 2304 > 11:48:47.374390 IP v20z-10.28902 > 172.25.41.159.2963: . ack 8237 win 1460 A couple of points. Linux doubles what you request so request x gets 2x, but then only 3/4 of that is available for real data, so receiver's advertised window winds up being 2*3/4x or 1.5x. And you also may need to set /proc/sys/net/ipv4/tcp_moderate_rcvbuf to 0 to disable Linux's autotuning feature. Some of this also may depend on the specific kernel version you're running. And the minimum setting from tcp_rmem may have an effect or may not be allowed to drop below the MTU size as Stephen pointed out. -Bill - To unsubscribe from this list: send the line "unsubscribe linux-net" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
