Port-xen archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: Dom0 bad network performance

Matthias Petermann <mp%petermann-it.de@localhost> writes:

> I have now run the test iperf3 with 128K window size on localhost
> again with netstat -s (before/after comparison). The system was
> rebooted before each test run and I ran the tests directly on the VGA
> console to exclude influences via SSH. When comparing the two diffs, I
> noticed:

Thanks.  Good move to do this fresh boot no network.

> - "packets sent" with Xen is only a third of the value without Xen. Is
> this an indication that different packet sizes are being used here,
> i.e. under Xen the packet size is larger by default?

As I understand it, iperf is testing for a specific time rather than a
specific amount of data.

> - "correct ACK header predictions" and "correct data packet header
> predictions" are many times higher without Xen than with Xen. Both
> parameters don't tell me much, but they sound like a high value would
> be beneficial here, i.e. another indication that something is wrong in
> the case with Xen?

I suspect that's not important, but it is interesting.  I see there was
one out-of-order packet without xen, and two with.  I didn't see
anything else in the diff that was concerning.

Jaromir suggested current; netbsd-9 XEN3_DOMU uses only 1 cpu, and my
understanding is that current runs SMP.  I'm guessing your box has a few
not-super-fast each cores, and so that might matter a lot.

Attachment: signature.asc
Description: PGP signature

Home | Main Index | Thread Index | Old Index