site_archiver@lists.apple.com Delivered-To: darwin-dev@lists.apple.com On Nov 29, 2006, at 1:18 , Anees Alappat wrote: Thanks & Regards Anees Good luck! -M _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.appl... Hi, We have our own implementation for linkAggregation.Currently it uses only 2 links to aggregate.The algorithm used is round robin.It simply switches alternative mbuf packets to each port.I expect this should improve a single tcp connection also.I would like to know the problem with using this distribution algorithm.Any way i m not getting any speed improvement. Could you please share you views on this case? You should try your physical NICs in a Linux box because of the numerous options available- that way, you will have a baseline against which you can compare. I have experience with round-robin channel bonding under Linux and it works well iff you can control the downstream trunk. Things to watch out for: 1) dropped packets/claimed collisions upstream (use udp iperf for software-detected drops) 2) dual-port NICs (such as from Intel) which are useless for bonding because they share interrupts (use bonding across physically-separate NICs only) 3) cheap gigabit switches that claim to handle trunking for bonding but still max out at/under 1GBps throughput To test the max throughput of your bonding implementation while eliminating as many factors as possible, it would be ideal to crossover bond first two Macs with your bond implementation and then your Mac and a Linux box for comparative purposes. This email sent to site_archiver@lists.apple.com