Re: Throughput deadlock

Rick Jones (raj@cup.hp.com)
Mon, 06 Jul 1998 10:59:00 -0700

The parameters you described seem to work fine for me on HP-UX 10.20.
10.20 is also "sort of" BSD and does not have TCP_NODELAY as a default -
nor does Solaris to the best of my knowledge.

# ./netperf -H hpntc11b -- -m 1024 -S 4096 -s 4096
TCP STREAM TEST to hpntc11b
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec

4096 4096 1024 10.00 8.49

rick jones

Edwin Law wrote:
>
> Hi,
> Has anyone tried using a send size of 1 KB and socket size of 4KB in
> an Netperf TCP_STREAM test? I got an exceptionally low throughput of
> 0.04Mbps. I believe it is caused by the so-called "throughput deadlock"
> problem, since 4KB < 3MSS and that turning on the TCP_NODELAY can
> eliminate this anomaly. Can anyone confirm this?
>
> How can sending more data (>1MSS) help solving the problem?
>
> Lastly, can anyone explain to me why this happens on IRIX (which is
> "sort of" BSD-based), but not on the STREAMS-based Solaris? Does Solaris
> has TCP_NODELAY set as default?
>
> Thanks in advance,
> Edwin.

-- 
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to email, or post, but please do not do both...
my email address is raj in the cup.hp.com domain...