Messages in this thread | | | Subject | RE: [PATCH] mm: moving dirty pages balancing to pdfludh entirely | Date | Tue, 4 Jul 2006 15:43:16 +0400 | From | "Ananiev, Leonid I" <> |
| |
Nikita Danilov wtites: >> Pdflush thread functions as before patching. Pdflush tends to make pages >> un-dirty without overload memory or IO and it is not need to let pdflush
> This assumption is valid for ext2
The assumption that pdflush should to make pages un-dirty without overload memory or IO is not for ext2 but for it sense. I'm working with ext3. A lot of work it does while writepages(). pdflush is throttled: while vmscan have sorted 32 page for paging-out it calls blk_congestion_wait() nevertheless had it put one of 32 page into congested queue or had not. pdflush is throttled.
Leonid
-----Original Message----- From: Nikita Danilov [mailto:nikita@clusterfs.com] Sent: Tuesday, July 04, 2006 1:55 PM To: Ananiev, Leonid I Cc: Linux Kernel Mailing List Subject: Re: [PATCH] mm: moving dirty pages balancing to pdfludh entirely
Ananiev, Leonid I writes: > Nikita Danilov wtites: > > performs page-out even if queue is congested. > Yes. If user thread which generates dirty pages need in > reclaimed memory it consider own dirty page as candidate for page-out. > It functions as before patching. > > > Intent of this is to throttle writers. > I suppose you means dirtier or write(2) caller but not writepage() > caller. The dirtier is throttled with backing_dev_info logic as before > patching.
I meant ->writepages() used by balance_dirty_pages(), see below.
> > While pdflush thread sorts pages for page-out it does not > consider as a candidate a page to be written with congested queue. > Pdflush thread functions as before patching. Pdflush tends to make pages > un-dirty without overload memory or IO and it is not need to let pdflush
This assumption is valid for ext2, where ->writepages() simply sends pages to the storage, but other file systems (like reiser4) do a *lot* of work in ->writepages() path, allocating quite an amount of memory before starting write-out. With your patch, this work is done from pdflush, and won't be throttled by may_write_to_queue() check, thus increasing a risk of allocation failure.
> do page-out with congested queue as you have proposed. > > Leonid
Nikita. - 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/
| |