[vortex] eth0: Too much work in interrupt, ...

[Hamid Reza Shahriari]

Dear friends,
thanks to your notices,

I should explain my benchmarks more. I have three machines
to benchmark running web browsers, proxy and web servers
(polyclt/proxy/polysrv). My benchmarker is Polygraph-2.2.9
(http://polyraph.ircache.net). polyclt simulates about 400
clients and each client generates request to proxy. 

You are right, I said that test terminated after 7 hours,
Its load was about 240/sec of PolyFill2 workload. Recent tests
used PolyMix2 workload, the former is used to filling the cache
and so its hit rate is very low, and it uses best effort load,
the later uses 55% hit rate and constant load.
The best effort load is easier for proxy.
The difference in crash time may be due to different workloads
or even by accident. I will try to repeat the tests.

Andrew,
'halt' or 'crash' in my mails means that I can't interact with
Linux by console, telnet and serial interface, so I must press
reset button. Can you tell me what does mean /proc/sys/vm/freepages
content? or where can I get document about it?

Now I'm using kernel 2.2.14 (don't ask why not 2.2.16, It was my
boss decision!). Proxy server eats the memory, and It is better
to increase the RAM.

Very Thanks 
--Hamid

Bogdan Costescu wrote:
>
> On Wed, 13 Sep 2000, Hamid Reza Shahriari wrote:
> 
> > Unfortunately, I used my all two motherboard mem slots by each 200MB. I
> > have not 512MB RAM chip. So I must buy a new 512MB chip.
> > It's 'free' output when testing:
> > # free
> >              total       used       free     shared    buffers     cached
> > Mem:        516436     513428       3008      13100     215288      44808
> > -/+ buffers/cache:     253332     263104
> > Swap:        72252       2176      70076
> 
> You seem to have quite a lot of memory (I asked because I supposed
> something like 64MB). But as 'free' shows, you use it all either for the
> applications or by the kernel buffering. Are you still using 2.2.10 ?
> Maybe upgrading to something newer will solve your problem; however, I
> don't have much knowledge in this area, so other oppinions are welcome!
> 
> > I means that error messages were reduced, and in high loads, it halts
> > later.
> 
> In one of your previous messages, you said that using 3c59x-bc-1.0 you
> were able to run for more than 7 hours, but with error messages. Now you
> say that it halts later and give a figure of 2 hours. Are you using the
> same driver ?
> 
> > The main bottleneck for me is kernel halt that is without any message when
> > halts! Is this due to disabling interrupts in NIC driver?
> 
> No, it's due to out-of-memory situation.
>