[netperf-talk] -s parameter doubt

[Rick Jones]

Frank Schuster wrote:
> Ok, but what value is used default for the sender and receiver buffer?
> My sysctl on ubuntu 9.10 with kernel 2.6.33 shows:
> net.sctp.sctp_mem = 47616	63488	95232
> net.sctp.sctp_rmem = 4096	277500	2031616
> net.sctp.sctp_wmem = 4096	16384	2031616
> this output.

My understanding from TCP experience is the SO_RCVBUF default is the 
middle value of rmem, and the SO_SNDBUF default is the wmem default:

net.ipv4.tcp_wmem = 4096        16384   2072576
net.ipv4.tcp_rmem = 4096        87380   2072576


raj at tardy:~/netperf2_trunk/src$ ./netperf
TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 
localhost.localdomain (127.0.0.1) port 0 AF_INET
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

  87380  16384  16384    10.00    2870.16


> netperf -H 192.168.1.101 -t SCTP_STREAM -l 1
> SCTP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.1.101 (192.168.1.101) port 0 AF_INET
> Recv   Send    Send                          
> Socket Socket  Message  Elapsed              
> Size   Size    Size     Time     Throughput  
> bytes  bytes   bytes    secs.    10^6bits/sec  
> 
> 233016 112640 112640    1.01       85.89
> 
> So, the recv socket is ok, because it is to another system and there
> is in sysctl exactly this value. But the send socket size 112640 Byte
> (the half is 61320) but I have no value for this in sysctl. How this
> value will be calculated?

That will require a Linux SCTP expert or a bit of kernel tree perusal to 
answer.

happy benchmarking,

rick jones

> 
> Regards
> Frank
> 
> 
>>Otavio Augusto wrote:
>>
>>>I was reading and test some netperf parameters and I'm doubt with the 
>>>'-s' or -S'.
>>>
>>>Example:
>>> netperf -H 127.0.0.1 -l 2 -- -s 10k -S 10k
>>>TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 127.0.0.1 
>>>(127.0.0.1) port 0 AF_INET
>>>Recv   Send    Send
>>>Socket Socket  Message  Elapsed
>>>Size   Size    Size     Time     Throughput
>>>bytes  bytes   bytes    secs.    10^6bits/sec
>>>
>>> 20000  20000  20000    2.00     1977.44
>>>
>>>
>>>Why it used the double? I set 10k but it use 20k?
>>>
>>>Thank You.
>>
>>Linux exposes the "overhead" for the socket buffer not just the actual 
>>byte count.  So, up to the limit of a couple sysctls, you will see a 
>>getsockopt() call return 2X what the setsockopt() call requested.  In 
>>other words, that is a "feature" of linux and is not being done by netperf.
>>
>>happy benchmarking,
>>
>>rick jones
> 
> ________________________________________________________________________
> Kostenlos tippen, täglich 1 Million gewinnen: zum WEB.DE MillionenKlick!
> http://produkte.web.de/go/08/