Messages in this thread | | | Date | Thu, 19 Mar 1998 17:06:57 +0100 | From | Alexander Kjeldaas <> | Subject | Re: [patch] fork() speedup, 2.1.90 |
| |
On Thu, Mar 19, 1998 at 02:29:40PM +0100, Jakub Jelinek wrote: > > > > > > The biggest overhead in current Linux's fork() latency is the creation > > of new page tables. This patch implements 'page table caching', a way to > > preserve constructed state after a process exits. Timings on a 100 MHz > > P5: > > > > Oh no, Ingo, save yourself the work and look sometimes into vger CVS. > Page table caching is implemented for all the platforms but m68k's pgd/pmd's > (just pte's are cached on m68k). It is tested and hopefully should go to the > mainstream during the upcoming merge. > But yes, it makes a big difference. > > There are some further things I want to do to speed up exit, but I'm waiting > for the merge - it would be mess otherwise. >
I'm implementing hashing of process groups and sessions in the normal pidhash table. This should speed up exit as well. We now have for_each_task_in_session, for_each_task_in_pgrp, find_task_by_pgrp, find_task_by_session and find_session_leader. Many uses of for_each_task can be updated to use these new functions.
astor
-- Alexander Kjeldaas, Guardian Networks AS, Trondheim, Norway http://www.guardian.no/
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu
| |