tech-kern archive
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]
Re: Weird memory usage, performance problem with 6.1_STABLE
> On 09/24/2013 04:31 AM, Mark Davies wrote:
>> On Fri, 20 Sep 2013, Mark Davies wrote:
>>> On Fri, 20 Sep 2013, Lars Heidieker wrote:
>>>> Can you see which kernel thread causes high CPU usage by showing
>>>> lwps in top? (t toggles those modes)
>>>
>>> 149 threads: 25 idle, 118 sleeping, 6 on CPU
>>> Memory: 15G Act, 15M Wired, 28M Exec, 15G File, 4620K Free
>>> Swap: 8192M Total, 8192M Free
>>>
>>> PID LID USERNAME PRI STATE TIME WCPU CPU NAME
>>> COMMAND
>>> 0 81 root 221 CPU/0 24:19 93.60% 93.60% pgdaemon
>>> [system]
>>> 13362 1 root 116 tstile/4 6:09 19.58% 19.58% -
>>> ftpd 17398 1 root 221 tstile/2 8:26 18.07% 18.07% -
>>> ftpd 14648 1 root 116 tstile/1 4:15 16.41% 16.41% -
>>> ftpd
>>>
>>>> or even better try:
>>>> systat vm 1
>>>> and check for page scan rates etc.
>>>
>>> When it was in the above state pdscn was reporting around 95000,
>>> pdfre was 0.
>>
>>
>> So is there something I can tune to get better behaviour out of this?
>> Just for comparison I tried the same thing on a 6.1_STABLE/i386 with
>> 4GB (3.3 available) and with 3 ftp's going it at sat at 2G active and
>> 1G inactive and remained responsive throughout the entire download.
>>
>> cheers
>> mark
>>
>
> I think the pagescanner can't keep up with the speed, is there any
> change in network bandwidth between those machines?
> Another idea, the images nearly fit into memory and the ftpd process
> trigger the file cache in a way that the page scanner can't find pages
> to free...
pagedaemon uses trylock to acquire vmobjlock.
it often fails if cache is occupied by a single active large file.
yamt-pagecache branch has some improvements in this area.
i don't know how i386 -> amd64 transition made the difference, though.
YAMAMOTO Takashi
> You wen't from non-smp to smp right?
>
> Lars
>
>
> --
> ------------------------------------
>
> Mystische Erkldrungen:
> Die mystischen Erkldrungen gelten f|r tief;
> die Wahrheit ist, dass sie noch nicht einmal oberfldchlich sind.
>
> -- Friedrich Nietzsche
> [ Die Frvhliche Wissenschaft Buch 3, 126 ]
Home |
Main Index |
Thread Index |
Old Index