Hi, Here are some values about the experiments. The values are the elapsed real time used by the process, in seconds. Each row corresponds to 10000 cat /proc/pid/smaps command.
Old smaps 19.41 19.31 21.38 20.16 New smaps 16.82 16.75 16.75 16.79 BR, Mauricio Lin. On Tue, 1 Mar 2005 10:17:56 -0400, Mauricio Lin <[EMAIL PROTECTED]> wrote: > Well, > > It is working better now. You are right Hugh. Now the new version is > faster than the old one. I removed the struct page and its related > function. > > Thanks, > > BR, > > Mauricio Lin. > > On Tue, 1 Mar 2005 04:08:15 -0400, Mauricio Lin <[EMAIL PROTECTED]> wrote: > > On Mon, 28 Feb 2005 20:41:31 +0000 (GMT), Hugh Dickins <[EMAIL PROTECTED]> > > wrote: > > > On Mon, 28 Feb 2005, Mauricio Lin wrote: > > > > > > > > Now I am testing with /proc/pid/smaps and the values are showing that > > > > the old one is faster than the new one. So I will keep using the old > > > > smaps version. > > > > > > Sorry, I don't have time for more than the briefest look. > > > > > > It appears that your old resident_mem_size method is just checking > > > pte_present, whereas your new smaps_pte_range method is also doing > > > pte_page (yet no prior check for pfn_valid: wrong) and checking > > > !PageReserved i.e. accessing the struct page corresponding to each > > > pte. So it's not a fair comparison, your new method is accessing > > > many more cachelines than your old method. > > > > > > Though it's correct to check pfn_valid and !PageReserved to get the > > > same total rss as would be reported elsewhere, I'd suggest that it's > > > really not worth the overhead of those struct page accesses: just > > > stick with the pte_present test. > > So, I can remove the PageReserved macro without no problems, right? > > > > > > > > > > Your smaps_pte_range is missing pte_unmap? > > Yes, but I already fixed this problem. Paul Mundt has checked the > > unmap missing. > > > > Thanks, > > > > Let me perform new experiments now. > > > > BR, > > > > Mauricio Lin. > > > - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
