> I can confirm that masa's binary fixes the bug. The > interface would die out after an hour of concurrent > file transfers. After changing the driver, it run > through the night finishing 130 GB of transfers each > way.
Hi, Thank you very much for testing. > Unfortunately, cifs read througput with this binary > is limited to ~12 MB/sec or half of what I can get > through the original rge driver (or anIntel desktop > 1000MT PCI card) on the same machine and the same > filesystem. > > It is possible that my CPU is running out of steam. I > am running snv_126 on a Supermicro Atom 330 > motherboard. With either lan controllers, one CPU > thread is maxed out. Masa, is your binary compiled > with debug assertions or anything else that would > limit performance? I think we unfortunately hit bug id 6888015, that pcie variants of realtek 8168/8111 series are very slow in transmitting. It is caused by the workaround for the hang issue. :-( I made a workaround for the slow down issue, but I have not find a sponsor to put back the fix into opensolaris. I'll prepare the binary of rge against the slow down issue soon. > BTW, using msi_enable=0 or replacing rge with the > gani driver plunged my throughput to less than 1Mbps. Oh, I didn't know the issue of gani. I just found gani had an issue to treat msi interrupts. I think I can fix it soon. regards, -masa > > annis -- This message posted from opensolaris.org _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org
