Okay, after I Did Stuff(tm) to my new computer, it's capable of
generating approximately 50Kints/s, which transates to 66MB/s in a
network benchmark (Actually, netio misreports the speeds, probably
because of an overflow; I was amused)
There's two howevers and one unfortunately, though.
The howevers:
- it's not in the direction I want -- meaning I should update, which I
can live with. But:
- the bench speed going from my old computer wend down from ~33mb to
~20MB/s which I can't make any sense of at all, and the practical
speeds from the server didn't change for the better, possibly for the
worse. I'm not sure, because of...
The unfortunately:
- After a few minutes, my old computer kernel paniced. I assume this
is because it can't handle the incoming traffic, as its own pace of
sending (and therefore probably handling interrupts) is more like
20kints/sec than 50. I couldn't tell anything vmstat-wise as during
the data going in that direction (un udp; tcp hung around 17MB both
ways - which is better thn before, but *still* short of impressive),
the ssh shells just stopped responding, and I guess so did vmstat.
It's probably being battered in hardware interrupts, I saw high
figures. Quite possibly this has to do with renicing ksoftircq
(negatively) as someone suggested. However, I am still of the opinion
that a difference in processor speeds should not crash anything.
The panic looks a lot like the last one; same kernel (napi still
enabled for the 8169). Image attached.
I'ld like to know what's wrong here, so that I can avoid it. I can
deal and live with stupid speeds better than crashing serves. Makes me
grumble less, y'know.
--Bart Alewijnse
panic2.gif
Description: GIF image
|
