[lkml]   [2014]   [Jun]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: fs/stat: Reduce memory requirements for stat_open
On 12.06.2014 15:41, Heiko Carstens wrote:
> On Thu, Jun 12, 2014 at 03:00:17PM +0200, Stefan Bader wrote:
>> When reading from /proc/stat we allocate a large buffer to maximise
>> the chances of the results being from a single run and thus internally
>> consistent. This currently is sized at 128 * num_possible_cpus() which,
>> in the face of kernels sized to handle large configurations (256 cpus
>> plus), results in the buffer being an order-4 allocation or more.
>> When system memory becomes fragmented these cannot be guarenteed, leading
>> to read failures due to allocation failures.
>> There seem to be two issues in play here. Firstly the allocation is
>> going to be vastly over sized in the common case, as we only consume the
>> buffer based on the num_online_cpus(). Secondly, regardless of size we
>> should not be requiring allocations greater than PAGE_ALLOC_COSTLY_ORDER
>> as allocations above this order are significantly more likely to fail.
>> The following patch addesses both of these issues. Does that make sense
>> generally? It seemed to stop top complaining wildly for the reporter
>> at least.
> Hi Stefan,
> see also

Hi Heiko,

doh, so you guys have been hit by that before. And I have missed the fact that
single_open is special. Which makes the change for the upper limit do the wrong
thing. While long-term it sounds like changing it to vmalloc or iterative reads
sounds better, maybe the change from possible to online cpus might be something
that is better acceptable as a quick fix... Not sure. At least this giving back
a bit of attention to the matter and it is not only affecting zSeries. x86
starts to see hw that requires a similar high possible cpus... :)


> and one possible solution:
> and the other one:
> Thanks,
> Heiko

[unhandled content-type:application/pgp-signature]
 \ /
  Last update: 2014-06-12 17:01    [W:0.061 / U:3.640 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site