Re: TCP vs. UDP on SMP

[julian_satran@il.ibm.com]

My experience is similar.  And this was also the reason that I stated that
using
several TCP connections would be beneficial to SMP machines (easing both
coupling and some serialization inherent to some stacks).

Julo

"Franco Travostino" <travos@nortelnetworks.com> on 18/09/2000 05:20:10

Please respond to "Franco Travostino" <travos@nortelnetworks.com>

To:   Black_David@emc.com, cslee@FalconStor.com.tw, John Hufferd/San
      Jose/IBM@IBMUS
cc:   ips@ece.cmu.edu (bcc: Julian Satran/Haifa/IBM)
Subject:  Re: TCP vs. UDP on SMP





>   My recollection of published results on using multiple
>processors for
>TCP in Unix-like operating systems is that the big gains come from using
>different
>processors for different connections rather than handling send and receive
>on a single
>connection on different processors due in part to the send/receive
coupling
>required
>by congestion control ... but it's been a while since I've checked/read
this
>literature.

Much agreed. Some quantitative results in:

Performance Issues in Parallelized Network Protocols, Erich M. Nahum, David
J. Yates, James
  F. Kurose, and Don Towsley. USENIX Symposium on Operating Systems Design
and
  Implementation (OSDI), Nov., 1994 (Monterey, CA).

Now that we're in TCP/IP offload age ... Recent programmable network
processors throw more silicon (e.g., hardware threads) and expose lower
level of control (e.g., asynchronous memory operations) than
general-purpose CPUs in SMP boxes. Kind of an SMP on a single chip, really.
While the TCP vs. UDP  fundamental parallelization issues are still in the
way, I expect that there will be a bunch of parallelization trade-offs more
when optimizing TCP on network processors.

-franco