Christo Kutrovsky wrote: > * Updated information. > > Apparently this is known issue with *all* Intel NICs. All "e1000" series. > Some newer ones do not have this issue. > > The issue is related with "write" speed over an Intel network *only* for TCP. > UDP is running at wire speed. "Read" from intel NIC is also running at wire > speed (114 Mbytes/sec) > This is both under opensolaris and linux.
Have you tried disabling the "hardware accelerator" cruft? I've run into no end of trouble with the various checksum and LSO features -- on the systems I run, they generally seem to make things slower, at least when they're not just corrupting packets. Having TCP transmit run poorly while all else is fine is a key symptom of such problems. The accelerator logic targets the TCP transmit path. Try putting this in /kernel/drv/e1000g.conf and reloading the driver: tx_hcksum_enable=0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0; lso_enable=0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0; -- James Carlson 42.703N 71.076W <[email protected]> _______________________________________________ networking-discuss mailing list [email protected]
