Van Jacobson. Efficient Protocol Implementation. Notes from SIGCOMM '90 Tutorial on "Protocols for High-Speed Networks". 1990.  Home/Search   Document Not in Database   Summary   Related Articles

....are shown, using different mark symbols on the graph. The modified kernel (shown with square marks) slightly improves the MLFRR, and avoids livelock at higher input rates. 1 This is not such a radical idea; Van Jacobson had already used it as a way to improve end system TCP performance [8]. 13 ELIMINATING RECEIVE LIVELOCK IN AN INTERRUPT DRIVEN KERNEL 0 12000 2000 4000 6000 8000 10000 Input packet rate (pkts sec) 0 6000 1000 2000 3000 4000 5000 Output packet rate (pkts sec) Unmodified No polling Polling (quota = 5) Polling (no quota) Figure 6 3: Forwarding ....

....saturate an Ethernet, so measuring TCP throughput over Ethernet shows no effect. The technique of processing a received packet directly from the device driver to the TCP layer, without placing the packet on an IP level queue, was used by Van Jacobson specifically to improve TCP performance [8]. It should reduce the cost of receiving a packet, by avoiding the queue operations and any associated locking; it also should improve the latency of kernel tokernel interactions (such as TCP acknowledgements and NFS RPCs) The technique of polling the interfaces should not reduce end system ....

Van Jacobson. Efficient Protocol Implementation. Notes from SIGCOMM '90 Tutorial on "Protocols for High-Speed Networks". 1990.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC