Hi, I've asked this question a couple of times on the main kernel mailing list but got no responses, so I wonder if any of the guys here can offer any thoughts. I've got a system running a 1TB software RAID-5 using SATA drives, along with a 3c2000 gigabit adapter. All works fine under 2.6.7, however the gigabit performance is terrible. Establish a single connection to the system initially ( a second or so ) delivers good performance of ~15-20MB/s, however it rapidly drops off over 2-3 seconds and settles at about 3MB/s. It's also very "bursty" according to the network utilisation graphs. The truly weird bit is that establishing another connection to the server from any system and transferring a file then causes the transfer rate to go back up to the ~15-20MB/sec again for the duration of the multiple transfers. My though was that it had something to do with the number of interrupts happening on the system, but there#s nothing sharing the network card interrupt, and I've also tried turning the interrupt mitigation on ( at 2000/sec ) which doesn't appear to have made much difference to this behaviour at all. Turning off the local APIC and disabling ACPI affected the behaviour slightly, which makes me think it's got something to do with the way interrupts are handled by the sk98lin driver. The MTU is set to 1500 ( limited by the switch ), everything is working at 1000FD, I've tried upping the wmax and rmax to 512K and setting the RX and TX queues to 2000 in an effort to tune the gigabit a little. Does anybody have any thoughts as to how to proceed sorting this out? Cheers in advance, Eamonn -- Eamonn Hamilton <hamiltonea@uk-aberdeen.mail.saic.com> - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
