> * ...which is that sometimes sockets seem... slow, for want of a
> better way to put it. Specifically, sending mail is very slow now.
> Both mail clients and sendmail seem to use a LOT of CPU now when
> doing SMTP, whereas in the past I could not have even measured
> it for a single message. strace shows a lot of socket calls,
> some failing
I have had similar problems. They started sometime after 2.1.131ac11
and still exist now in vanilla 2.1.132. First I noticed X11 running
slower than a herd of bored pigs chasing tiger slugs. I have a 4Mb
Permedia 2 card, a slower-clocked box here with a 512K Trident is much
faster in X. Perhaps there's just something wrong with the
XF86_3DLabs driver (and I admit, I'm using the 3.3.3 server with my
old 3.3.2.3a everything-else-but-the-server, perhaps this is bad?) but
then that doesn't explain why smail at exactly the same time decided
that it couldn't bind sockets (returns "operation timed out"). The
only two changes to the system at the time was the new video card and
the new kernel. Strangely enough, I "fixed" the smail problem by
editing the DHCP-given /etc/resolv.conf setup and adding in 127.0.0.1
as the first nameserver entry (and bind is running on the localhost,
caching-only setup with the other DNS servers in resolv.conf as the
forwarders). Your guess is as good as mine as to why this fixed it -
both with and without it in, nslookup functions perfectly.
Oh, and TCP/IP is only slightly faster than X. I'm getting 157kB/s
_max_ - that was scp'ing (without compression) 80Mb worth of data from
the server box across the room, on the local ether. Normally I can
get at _least_ three times that out of my poor old PCI ne2k. The
_very_ strange thing was that the transfer rate didn't alter one iota
when I loaded up a far away remote web site, and did a few other
network-related things. The web site came down at a normal speed for
these parts, which was also surprising. It's like there's some new
speed cap on socket i/o. Speed might kill, but the dork sitting in the
fast lane going 10% of the speed limit is the one that needs arresting!
-- Matt- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/