Re: kernel memory allocators

[Lars Heidieker <lars.heidieker%googlemail.com@localhost>]

On Wed, Jan 19, 2011 at 6:02 PM, Allen Briggs <briggs%netbsd.org@localhost> 
wrote:
> On Wed, Jan 19, 2011 at 05:55:57PM +0100, Lars Heidieker wrote:
>> > What are the benefits that you expect to see from doing this work?
>> > Have you measured to see if those benefits are realized?
>>
>> Benefits I've thought about:
>> - The kmem pools use pool_caches therefor scalability will be much
>> better as the old malloc has a single lock for all access, the pools
>> have one each with a per cpu cache layer.
>> - The old malloc only returns oversized allocations back to the kmem
>> layer but nothing that is in it's bucket, pools can be drained...
>> - Removing one redundant interface in the kernel-api (in the long
>> term, when dropping the malloc wrapper)
>
> Fair enough.  Do you have any measurements to see if the benefits are
> realized?  Does it actually (not just theoretically) scale better?
> Does draining happen and the system behave better in low-memory
> situations?  Does it stay up when it would panic/crash/stall before?
> What is the code size savings for getting rid of the malloc code (and
> leaving a shim or not)?
>

Scalability should be better, but depending on how much the old malloc
was used, I don't think it will be much.
I am going to do the following measurements:
- total kernel memory consumption.
- count of kmem_map /kernel_map entries these should be down with the
extended virtual-address-caching at the kmem_map / kernel_map layer
(doing the equivalent to vmem quantum caches). fragmentation in those
maps should go down.
- memory size after running reclaims on the caches.

kernel size is down a bit (build from the same config (i386))
   text    data     bss     dec     hex filename
3214524   79288  257308 3551120  362f90 /netbsd
3209676   78488  258716 3546880  361f00 /netbsd.test


>> I am in the progress of some cleanup work, I'll make a patch available
>> within the next few days.
>
> Cool!
>
> -allen
>
> --
> Allen Briggs  -  briggs%ninthwonder.com@localhost
>



-- 
Mystische Erklärungen:
Die mystischen Erklärungen gelten für tief;
die Wahrheit ist, dass sie noch nicht einmal oberflächlich sind.
   -- Friedrich Nietzsche
   [ Die Fröhliche Wissenschaft Buch 3, 126 ]