ULE and very bad responsiveness

Hi,

from comp.unix.bsd.freebsd.misc:

Kris Kennaway wrote:

> On 2003-11-13, Harald Schmalzbauer <news_at_schmalzbauer.de> wrote:
> 
>> Well, I don't have any measurements but in my case it's not neccessary at
>> all. I built a UP kernel with ULE like Kris advised me.
> 
> Are you running an up-to-date 5.1-CURRENT?  ULE was broken with these
> characteristics until very recently.  If you're up-to-date and still
> see these problems, you need to post to the current mailing list.
> 
> Kris

Yes, I am running current as of 13. Nov.

Find attached my first problem description.

Thanks,

-Harry

attached mail follows:

Harald Schmalzbauer wrote:

> Kris Kennaway wrote:
> 
>> On 2003-11-12, Ivan Voras <ivoras_at__fer_.hr> wrote:
>>> Kris Kennaway wrote:
>>>> On 2003-11-11, Ivan Voras <ivoras_at__fer_.hr> wrote:
>>>>
>>>>> Debugging in 5-current halves the overall performance for non
>>>>> cpu-only processes and ULE scheduler still offers less performance
>>>>> than 4BSD.
>>>>
>>>> No, your data shows no statistically-significant difference between
>>>> ULE and 4BSD.
>>>
>>> Well, I agree 6 measurements[*] are not much, but then again, it points
>>> to a correlation and some folks need all the speed they can get (and
>>> don't have MP systems).
>>>
>>>
>>> [*] bytebench runs 6 measurements for each test.
>> 
>> There's no correlation unless you do enough measurements to measure
>> it.  Your numbers showed an tiny difference which may or may not be
>> real.
> 
> Well, I don't have any measurements but in my case it's not neccessary at
> all. I built a UP kernel with ULE like Kris advised me.
> Seti is always running with nice 15.
> When I try to install a already built (compiled) port it takes about 2
> minutes to complete (in this case the nvidia-driver).
> When I stop seti it is done in about 5 Seconds like usual. Also when
> building a port, the first make-fetch-extract-patch-configure steps take
> 100 times the time against seti not running.
> With the old scheduler I dind't even recognize that seti was running
> (that's what nice 15 should affect I think).
> On the other hand, starting up kde doesn't make any noticable difference,
> but starting (and working) with kmail does. It's not the factor 100 or so
> like with the make process but noticably more sluggish where I couldn't
> feel any difference with the old scheduler.
> 
> Please tell me what I should do. If there's no solution I'll switch back
> in a view days because I loved BSD for the great responsiveness under high
> load. I always had seti runnning, compiling in the backgroung, listening
> music and do my daily work without any problem on a 1Ghz machine.
> 
> Thanks,
> 
> -Harry
> 
> 
> 
>> 
>> Kris