Re: [libpcap] High latency on i7 vs. Core 2 Duo systems (#304)

[Michael Richardson <mcr () sandelman ca>]

> > > > > "sbourland" == sbourland  <notifications () github com> writes:
    sbourland> I have a very simple program that watches for packets on
    sbourland> an interface and then sends a reply packet.  On all the
    sbourland> Core 2 Duo systems I have tested, the response time is on
    sbourland> the order of <20 usec (measured using a separate machine
    sbourland> running tcpdump fed packets from a mirrored port).  When
    sbourland> attempting to upgrade to i7 based systems, the response
    sbourland> time is now over 100 usec.  Obviously Intel has been less
    sbourland> than interested in helping figure out what has happened.
    sbourland> Have tested with multiple Core 2 Duo and i7 systems with
    sbourland> consistent results.  I am hoping someone on here as

Identical kernels?
Identical network interface cards?

I would suspect that the difference of 80us is the result of different
mechanisms being used by the NIC card to either signal arrival of the
packet (interrupt mitigation mechanisms, for instance), and/or the same
thing on the transmit side.

-- 
]               Never tell me the odds!                 | ipv6 mesh networks [ 
]   Michael Richardson, Sandelman Software Works        | network architect  [ 
]     mcr () sandelman ca  http://www.sandelman.ca/        |   ruby on rails    [ 
        

_______________________________________________
tcpdump-workers mailing list
tcpdump-workers () lists tcpdump org
https://lists.sandelman.ca/mailman/listinfo/tcpdump-workers