My end user testing of 2.4.8-ish kernels

My end user testing of 2.4.8-ish kernels

[J Sloan]

I did some testing today on the mini-low-latency patch.

I must admit that I was totally biased towards it from the start.

While it certainly didn't hurt anything, the bottom line is that
after hours of mp3/dbench tests, I was unable to quantify any real
difference between 2.4.18-pre3 vanilla and with mini low latency.
They exhibit pretty much the same behaviour in terms of how much
dbench it takes to start hearing audio dropouts in xmms - they were
both smooth up to dbench 40, but started exhibiting sporadic audio
dropouts at dbench 64.

Out of curiosity I booted up 2.4.18pre2-aa2 and found it a real gem.
To my pleasant suprise I was able to run dbench 128 without hearing
a _single_ audio dropout. (the dbench 128 result was 19.75 MB/sec)

With dbench 192 I did start to hear some occasional dropouts, but
they were generally short, e.g. 100ms or so.

In any event, all the 2.4.18-pre-ish kernels I tested today are much
better at this than e.g. 2.4.7 - at least on my hardware, I am now
getting excellent interactive performance under load without preempt
or low-latency patches, and that's a good thing.

IMHO the -aa kernel seems to the clear winner here -

Good for server use, good for desktop use...

Regards

jjs

Re: My end user testing of 2.4.8-ish kernels

[Andrew Morton]

J Sloan wrote:
> 
> I did some testing today on the mini-low-latency patch.
> 
> I must admit that I was totally biased towards it from the start.
> 
> While it certainly didn't hurt anything, the bottom line is that
> after hours of mp3/dbench tests, I was unable to quantify any real
> difference between 2.4.18-pre3 vanilla and with mini low latency.
> They exhibit pretty much the same behaviour in terms of how much
> dbench it takes to start hearing audio dropouts in xmms - they were
> both smooth up to dbench 40, but started exhibiting sporadic audio
> dropouts at dbench 64.

Oh well.   I must have missed one.

> Out of curiosity I booted up 2.4.18pre2-aa2 and found it a real gem.
> To my pleasant suprise I was able to run dbench 128 without hearing
> a _single_ audio dropout. (the dbench 128 result was 19.75 MB/sec)
> 
> With dbench 192 I did start to hear some occasional dropouts, but
> they were generally short, e.g. 100ms or so.
> 
> In any event, all the 2.4.18-pre-ish kernels I tested today are much
> better at this than e.g. 2.4.7 - at least on my hardware, I am now
> getting excellent interactive performance under load without preempt
> or low-latency patches, and that's a good thing.
> 
> IMHO the -aa kernel seems to the clear winner here -
> 

the -aa kernel basically includes everything that's in the mini-ll
patch.  If you merge -aa, you get mini-ll.  Plus the one I missed :)

-

Re: My end user testing of 2.4.8-ish kernels

[J Sloan]

Andrew Morton wrote:

>the -aa kernel basically includes everything that's in the mini-ll patch.  If you merge -aa, you get mini-ll.  Plus the one I missed :)
>
Ah, that explains a lot -

OK, things are making more sense now...

I look forward to mini-ll and the other tweaks
going into the mainstream kernel!
 
cu

jjs