Time taken for send/receive.

Hi All,

Does anybody know of any study conducted to benchmark the linux
networking stack?

Typically does it take more instructions to process an incoming packet
when compared to sending a packet?

Eg. If an application does a send and if we are studying the processing
below the TCP/UDP layer, till the packet is given the NIC drivers, does
it take more time to send a packet than to process an incoming one?

Its well known that encoding takes more time than decoding, what about
send and receive of a particular IP packet?

Any pointers on this will be highly appreciated.

Thanks,
Ajay

(E-Mail Removed) wrote:
> Does anybody know of any study conducted to benchmark the linux
> networking stack?

Has anyone ever measured the length of a piece of string?-)

> Typically does it take more instructions to process an incoming
> packet when compared to sending a packet?

You first must define typical. With CKO in the NIC, and copy
avoidance, it takes fewer cycles to send a KB of data than to receive.
Add TSO (large send) to that mix and it shifts even further.

> Eg. If an application does a send and if we are studying the processing
> below the TCP/UDP layer, till the packet is given the NIC drivers, does
> it take more time to send a packet than to process an incoming one?

q-syscollect (www.hpl.hp.com) and/or oprofile would seem to be your
friend here.

rick jones
--
oxymoron n, Hummer H2 with California Save Our Coasts and Oceans plates
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...