On Sunday 23 October 2005 06:18, dileep reddy wrote: > Hi All, > I plan to change tcp code in the kernel(2.6.13.2) to > include a custom algorithm for congestion control. ] Consider using a congestion control module. "Pluggable congestion avoidance modules" implementation recently included gives you the interface to build such a module. > My work is an extension to the Reno algorithm. So, I dont > want to use any of the sack, fack, or dsacks. > Will disabling the parameters in /proc/sys/net/ipv4 > directory would suffice? Yes. SACK will only work if it is enabled on both sides. > 3. Also, in the same file, I found the comment > "emulate sack for sackless connection" at the function > tcp_add_reno_sack(). Does that mean that even after > disabling sack, the estimate about sacked packets > would be made? It could be that linux behavior when recovering from errors isn't much close to what you are expecting. I'd suggest you to re-read tcp_input.c, caring more about rate-halving and ECN/SACK/FACK state machine. Also, I think that [*] contains your answers. (IMHO it is still one of the best papers that describes the basics of SACK/DSACK/FACK implementation). [*] == P. Sarolahti and A. Kuznetsov. "Congestion Control in Linux TCP". In Proceedings of Usenix 2002. Monterey, CA, USA, June 2002. -- Daniele Lacamera root at danielinux.net - : 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
