Re: 15000+ processes -- poor performance ?!

William Lee Irwin III (wli@holomorphy.com)
Wed, 18 Dec 2002 18:01:47 -0800


On Thu, 2002-12-19 at 01:04, Perez-Gonzalez, Inaky wrote:
>> If it has it ... well, I have no idea - maybe Robert Love would know.

On Thu, Dec 19, 2002 at 02:31:28AM +0000, Alan Cox wrote:
> He's running the -aa kernel, which has all the right bits for this too.
> In fact in some ways for very large memory boxes its probably the better
> variant

In my experience the most critical issues running 16K processes are:
(1) the highmem footprint of the pte's is significant
(2) the lowmem footprint of pmd's

and most of the rest is in the noise. It's probably a bad idea to run
top(1) or perhaps even mount /proc/ at all until top itself,
proc_pid_readdir(), and the tasklist_lock are all fixed.

Pretty much all he needs to "stay alive" is highpte of some flavor or
another. Performance etc. is addressed somewhat more by 2.5.x than -aa,
at least in the context of not degrading with this kind of multitasking.
i.e. shpte and pidhash. I've been randomly shooting down do_each_thread()
and for_each_process() loops in -wli, which is why I recommended it.

Bill
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/