[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]
Re: Dom0 bad network performance
Am 13.02.21 um 22:14 schrieb Jaromír Doleček:
There is one other thing that I remembered from last time I was looking at this.
IIRC it's just the network performance Dom0 <-> outside which was slow
compared to native, for DomU (running on that Dom0) <-> outside it was
line-speed nevertheless for me.
Perhaps then this is a red herring, and not really worth investigating
- you will not really be running anything directly on the Dom0.
If that were the case, then I would be able to live with it because I
see it the same way as you - in Dom0, the network speed is not so
important to me. However, even in the DomU's no better performance is
retrievable - regardless of whether the DomU is a NetBSD 9.1 stable or a
NetBSD current. My comparison measurement with iperf brought it to
almost identical values around 25 MB/s for NetBSD 9.1 Stable. The values
for NetBSD-current were noticeably lower at ~18 MB/s.
peterma@nuc:~> iperf3 -w 128K -c 192.168.2.51
Connecting to host 192.168.2.51, port 5201
[ 5] local 192.168.2.40 port 35990 connected to 192.168.2.51 port 5201
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-1.00 sec 25.5 MBytes 214 Mbits/sec 0 164 KBytes
[ 5] 1.00-2.00 sec 26.8 MBytes 225 Mbits/sec 7 192 KBytes
[ 5] 2.00-3.00 sec 26.3 MBytes 221 Mbits/sec 26 151 KBytes
> There are a bunch of optimizations particularly in pmap, which are
> disabled on Dom0 kernel compared to the native one. It also needs to
> do MMU updates via Xen hypercalls. So it will always be somewhat
> slower than native.
Well, maybe my combination of low-end CPU and NetBSD 9.1 Dom0 without
SMP is actually the decisive factor that makes the lack of optimizations
particularly noticeable here. I still have a little hope that maybe it's
just a setting thing in the network stack? During the test, the CPU is
80% idle, so would have reserves.
Main Index |
Thread Index |