[ntp:hackers] NIC knacks

Poul-Henning Kamp phk at phk.freebsd.dk
Sun Feb 10 21:13:49 UTC 2008


In message <47AF67BD.1040504 at udel.edu>, "David L. Mills" writes:

>Here is a snip from a memo I am writing. I read it again and am puzzled 
>about the apparent conclusion that the incidental delay of each 
>transceiver - transmit and receive - is 50 us. That sound really high 
>and implies a huge number of machine cycles. Is there a driver/NIC 
>expert onboard that has insight?

50 usec is not at all an unreasonable number for "modern" hardware,
there is far too much gunk to set up and far too much shared memory
traffic to get quick reactions these days.

On top of it, many nics will intentionally delay interrupts in the
hope that each interrupt will then serve multiple packets.

Which particular network hardware/drivers are you measuring this on ?

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.


More information about the hackers mailing list