Date: Thu, 18 Feb 1999 22:11:55 -0600 From: Jonathan Lemon <jlemon@americantv.com> To: freebsd-hackers@freebsd.org Subject: Ethernet interrupt overhead Message-ID: <19990218221154.34584@right.PCS>
next in thread | raw e-mail | index | archive | help
I have an application which is supposed to handle a lot of
ethernet traffic (a proxy web server, actually). However,
when under load, the kernel seems to spend an inordinate amount
of time handling network interrupts.
The machine in question is a P-II 300 with ~400MB memory, and
is connected to a private 100BX switch. When using either of
the following adapters:
xl0: <3Com 3c905 Fast Etherlink XL 10/100BaseTX>
rev 0x00 int a irq 10 on pci0.11.0
xl0: autoneg complete, link status good (full-duplex, 100Mbps)
or
pn0: <82c168/82c169 PNIC 10/100BaseTX> rev 0x21 int a irq 10 on pci0.11.0
pn0: autoneg complete, link status good (full-duplex, 100Mbps)
I'm seeing (as reported via systat) that the machine is spending
about 30% of it's time handling interrupts. The ethernet card is
generating just under 10,000 interrupts per second.
This seems to translate into roughly 9,000 cycles/packet, which
seems rather high to me. Is this reasonable, or do I just have
lousy ethernet cards? Would the EtherExpress (fxp0 driver) perform
better under this load?
--
Jonathan
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19990218221154.34584>