arca-vm-26

arca-vm-26

[Andrea Arcangeli]

I've put out an arca-vm-26. The differences between arca-vm-25 and -26 are
that I don't do aging on the swap cache anymore (this should allow
shrink_mmap to decrease the size of the cache at least more easily also on
low memory systems), and a modifyed mem trashing heuristic. It fix also a
potential swap deadlock (triggerable only by the shm memory I think).

The patch can be downloaded from:

ftp://e-mind.com/pub/linux/kernel-patches/2.2.0-pre8testing-arca-VM-26.gz

If you try it please feedback as usual ;). thanks.

Andrea Arcangeli

--
This is a majordomo managed list.  To unsubscribe, send a message with
the body 'unsubscribe linux-mm me@address' to: majordomo@kvack.org

Re: arca-vm-26

[Andrea Arcangeli]

On Wed, 20 Jan 1999, Andrea Arcangeli wrote:

> I've put out an arca-vm-26. The differences between arca-vm-25 and -26 are

I've released now arca-vm-27. I would like to know if you see
some difference between arca-vm-26 and -27. The main difference is that I
removed again the swapout smart code (that I inserted at PG_dirty time
trying to decrease the stalls caused by the swap code that was running
too much).

ftp://e-mind.com/pub/linux/kernel-patches/2.2.0-pre8testing-arca-VM-27.gz

Thanks.

Andrea Arcangeli

--
This is a majordomo managed list.  To unsubscribe, send a message with
the body 'unsubscribe linux-mm me@address' to: majordomo@kvack.org

[patch] arca-vm-28 - new nr_freeable_pages

[Andrea Arcangeli]

I have a new arca-vm-28. I don't have time to comment the changes right
now, but I would like if you could try it and feedback. This is not
intended to be good in low memory system, it _could_ work fine also with
low mem, but I don't know...

I can tell you that the bench that was running in 100 sec in pre1 and in
50 sec since I killed kswapd, now it runs in 20 sec. Iteractive
performances seems good. Let me know.

Forget to tell, if it will be crazy or it will not compile, I could have
done mistakes in diffing the stuff (it's becoming a relevant work ;), in
such case you can download arca-109 that has arca-vm-28 included as well. 

ftp://e-mind.com/pub/linux/kernel-patches/2.2.0-pre8testing-arca-VM-28.gz

Andrea Arcangeli

--
This is a majordomo managed list.  To unsubscribe, send a message with
the body 'unsubscribe linux-mm me@address' to: majordomo@kvack.org

Re: [patch] arca-vm-28 - new nr_freeable_pages

[Andrea Arcangeli]

On Thu, 21 Jan 1999, Andrea Arcangeli wrote:

> ftp://e-mind.com/pub/linux/kernel-patches/2.2.0-pre8testing-arca-VM-28.gz

Woops I forgot to diff pagemap.h in arca-vm-28, excuse me. At least
arca-tree-109 was fine ;). 

Index: linux/include/linux/pagemap.h
diff -u linux/include/linux/pagemap.h:1.1.1.1 linux/include/linux/pagemap.h:1.1.2.4
--- linux/include/linux/pagemap.h:1.1.1.1	Mon Jan 18 02:27:10 1999
+++ linux/include/linux/pagemap.h	Fri Jan 22 10:12:47 1999
@@ -11,17 +11,16 @@
 
 #include <linux/mm.h>
 #include <linux/fs.h>
+#include <linux/swap.h>
 
 static inline unsigned long page_address(struct page * page)
 {
-	return PAGE_OFFSET + PAGE_SIZE * page->map_nr;
+	return PAGE_OFFSET + (page->map_nr << PAGE_SHIFT);
 }
 
-#define PAGE_HASH_BITS 11
+#define PAGE_HASH_BITS 13
 #define PAGE_HASH_SIZE (1 << PAGE_HASH_BITS)
 
-#define PAGE_AGE_VALUE 16
-
 extern unsigned long page_cache_size; /* # of pages currently in the hash table */
 extern struct page * page_hash_table[PAGE_HASH_SIZE];
 
@@ -58,7 +57,7 @@
 			break;
 	}
 	/* Found the page. */
-	atomic_inc(&page->count);
+	page_get(page);
 	set_bit(PG_referenced, &page->flags);
 not_found:
 	return page;
@@ -78,6 +77,7 @@
 		page->pprev_hash = NULL;
 	}
 	page_cache_size--;
+	page_put(page);
 }
 
 static inline void __add_page_to_hash_queue(struct page * page, struct page **p)


Andrea Arcangeli

--
This is a majordomo managed list.  To unsubscribe, send a message with
the body 'unsubscribe linux-mm me@address' to: majordomo@kvack.org

Re: [patch] arca-vm-28 - new nr_freeable_pages

[Andrea Arcangeli]

On Thu, 21 Jan 1999, Andrea Arcangeli wrote:

> I have a new arca-vm-28. I don't have time to comment the changes right
> now, but I would like if you could try it and feedback. This is not
> intended to be good in low memory system, it _could_ work fine also with
> low mem, but I don't know...

I've done some further changes. nr_freeable_pages is still there (since
worked fine so far). But the new code should be far more friendly in low
memory. I am doing everything in 32 Mbyte (as some month ago by default
btw ;) to force me to test the new code ;).

This new code will not give a swapout raw speed as the previous one but
looks far more sane for low memory conditions. The old code was extreme,
the best to half the swapout time and to get the best numbers, but too
much aggressive to use the machine for doing other tasks at the same time
in low memory.

I would like to hear comments if somebody will try it. If you care about
low memory machines please try it and let me know.

ftp://e-mind.com/pub/linux/arca-tree/2.2.0-pre9_arca-3.gz

Now it's really time to sleep for me...

Ah forget to tell, if the size of the cache looks too high or too low,
feel free to decrease or increase the first value of /proc/sys/vm/pager

--
To unsubscribe, send a message with 'unsubscribe linux-mm my@address'
in the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://humbolt.geo.uu.nl/Linux-MM/