You should test further. I tried to reproduce your results and found =
that initially things look this way. But if I ssh into the colinux =
instance and try the tests, then the colinux' netperf network =
performance is actually better than loopback. =20
Even when I then switch back to using the default CoLinux console after =
this, the netperf network results are as they were in the ssh session. =20
So maybe you need at least 1 open TCP session to achieve acceptable =
performance on this test. Try ssh-ing from one instance to the other =
and then run the test from that one to the first. And then reverse this =
procedure.
I'm using the TAP bridging scheme in XP. How are you setup?
--=20
--Blake
pb...@ta...
"Taisuke Yamada" <tya...@li...> wrote in message =
news:nqt...@ho......
> Hi.
>=20
> I've set up 2 coLinux instances with NFS client-server
> configuration to save diskspace, but performance is way
> slower than I expected.
>=20
> To quantify my experience, I dis some benchmarking with
> netperf. In following data, "co00" is the NFS server
> and "co01" is the client. They both use TAP driver bridged
> by Windows "Network Bridge" for network connection.
>=20
> co00:/root# netperf
> TCP STREAM TEST to localhost
> Recv Send Send
> Socket Socket Message Elapsed
> Size Size Size Time Throughput
> bytes bytes bytes secs. 10^6bits/sec
> =20
> 87380 16384 16384 10.00 8599.03
> co00:/root#
>=20
> co01:/root# netperf -H 192.168.100.100
> TCP STREAM TEST to 192.168.100.100
> Recv Send Send
> Socket Socket Message Elapsed
> Size Size Size Time Throughput
> bytes bytes bytes secs. 10^6bits/sec
> =20
> 87380 16384 16384 10.02 23.39
> co01:/root#
>=20
> As you can see, TCP thoughput between 2 colinux instances
> is about 400 times slower than loopback.
>=20
> Is this a current limitation, or am I doing something wrong?
> Since there's no physical device between these two, I'm
> wondering if there's any coLinux parameter to increase
> network performance.
>=20
> Thanks in advance.
>=20
|
