[netperf-talk] meaning of throughput in netperf results
Rick Jones
rick.jones2 at hp.com
Mon Nov 22 10:31:19 PST 2010
bored to death wrote:
> hi everybody,
>
> i wanted to perform standard performance benchmarking on my router based
> on RFC2544 (Benchmarking Methodology for Network Interconnect Devices).
> in this RFC (and RFC1242 which is used by this), the meaning of
> "throughput" is defined:
>
> "Throughput: The maximum rate at which none of the offered frames are
> dropped by the device"
>
> meaning, throughput is the maximum bandwidth in which frame-loss is
> exactly zero.
>
> so i wanted to know:
> 1- is "netperf"'s throughput result, based on this definition or not
> (for example, if netperf says, throughput is 500Mbps, does it mean we
> have zero frame-loss for traffic up to 500Mbps)?
> 2- if so, does using netperf with regular parameters, work for me? for
> example, do these commands, give me what i want (throughput with
> RFC-based definition, for TCP and UDP streams)?
> # netperf -c -l 60 -H server -t TCP_STREAM -i 10,2 -I 99,5 -- -m 64
> -s 57344 -S 57344
The TCP_STREAM test does not know if there were any packet losses. TCP will
have retransmitted and the data will have arrived. There may be hiccups in the
length of time spent in a send() call or something, but within the confines of
netperf itself, you will not "know" that the test completed without any packet
loss. If you also check things like netstat statistics, then you can know if
there were any TCP retransmissions and so presumably packet losses.
> # netperf -c -l 60 -H server -t UDP_STREAM -i 10,2 -I 99,5 -- -m 64
> -s 57344 -S 57344
As the docs point-out, there will be two lines in the output for this test. The
first line will be what the sender thought it did. The second line will be what
the receiver actually received. So, if sends == receives then you know there
was no packet loss during that test.
happy benchmarking,
rick jones
>
> 3- if netperf can't do this, does anybody have any suggestions for me?
> is there any other tools that can help me?
More information about the netperf-talk
mailing list