[Etherboot-discuss] iSCSI slowness
Brought to you by:
marty_connor,
stefanhajnoczi
From: Thomas M. <tho...@gm...> - 2009-06-17 11:45:00
|
Hi everyone, I'm experiencing a slowness when I iSCSI boot DOS and run scandisk. I'm using the latest gPXE-git version and was able to reproduce the issue with rtl8139, r8169 and eepro100 cards in the client. I was using a D-Link DGE-530T gigabit card in the server(linux driver is skge). The 2 machines were directly connected using a Cat5e 4-pair crossover cable. Using a via-rhine card in the server or keeping the skge card, but putting a switch in between those two PCs fixes the problem. When experiencing the slowness, it seems that scandisk is stuck for a short moment and continues again. It gets stuck about every other second. Usually I'd say it's some kind of hardware problem, because changing the server network card or using a switch fixes the problem. I'm writing to the list anyway because HTTP transfers with the same setup(skge card in the server, same crossover cable, no switch, any of the 3 mentioned cards in the client) are really fast. I usually get around 11MB/sec for HTTP transfers(measured with the time command in gPXE). mdc seemed to experience similar(the same?) problems also when testing some driver changes I made. As far as I know he was using a switch, and both the server and client network cards were eepro100 cards. I'm using IET as iscsi target, iscsid version 0.4.16 I have two packet captures, one for the slow transfer and one for the fast transfer at http://etherboot.org/share/meteger/dumps.tar.gz Please let me know if you need any additional information. Thanks in advance for any answers Thomas Miletich |