Patchwork [PATCHv3,7/9] migration: do not sent zero pages in bulk stage

login
register
mail settings
Submitter Peter Lieven
Date March 21, 2013, 3:57 p.m.
Message ID <1363881457-14814-8-git-send-email-pl@kamp.de>
Download mbox | patch
Permalink /patch/229743/
State New
Headers show

Comments

Peter Lieven - March 21, 2013, 3:57 p.m.
during bulk stage of ram migration if a page is a
zero page do not send it at all.
the memory at the destination reads as zero anyway.

even if there is an madvise with QEMU_MADV_DONTNEED
at the target upon receipt of a zero page I have observed
that the target starts swapping if the memory is overcommitted.
it seems that the pages are dropped asynchronously.

Signed-off-by: Peter Lieven <pl@kamp.de>
---
 arch_init.c |   10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
Eric Blake - March 21, 2013, 7:26 p.m.
On 03/21/2013 09:57 AM, Peter Lieven wrote:
> during bulk stage of ram migration if a page is a
> zero page do not send it at all.
> the memory at the destination reads as zero anyway.
> 
> even if there is an madvise with QEMU_MADV_DONTNEED
> at the target upon receipt of a zero page I have observed
> that the target starts swapping if the memory is overcommitted.
> it seems that the pages are dropped asynchronously.
> 
> Signed-off-by: Peter Lieven <pl@kamp.de>
> ---
>  arch_init.c |   10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)

>              if (is_zero_page(p)) {
>                  acct_info.dup_pages++;
> -                bytes_sent = save_block_hdr(f, block, offset, cont,
> -                                            RAM_SAVE_FLAG_COMPRESS);
> -                qemu_put_byte(f, *p);
> -                bytes_sent += 1;
> +                if (!ram_bulk_stage) {
> +                    bytes_sent = save_block_hdr(f, block, offset, cont,
> +                                                RAM_SAVE_FLAG_COMPRESS);
> +                    qemu_put_byte(f, 0);
> +                }
> +                bytes_sent++;

Logic is STILL wrong.  I pointed out in v2 that bytes_sent should not be
incremented if you are not sending the page, so it needs to be inside
the 'if (!ram_bulk_stage)'.

Do we want to add a new migration statistic counter of how many zero
pages we omitted sending during the bulk stage?
Peter Lieven - March 21, 2013, 7:44 p.m.
Am 21.03.2013 um 20:26 schrieb Eric Blake <eblake@redhat.com>:

> On 03/21/2013 09:57 AM, Peter Lieven wrote:
>> during bulk stage of ram migration if a page is a
>> zero page do not send it at all.
>> the memory at the destination reads as zero anyway.
>> 
>> even if there is an madvise with QEMU_MADV_DONTNEED
>> at the target upon receipt of a zero page I have observed
>> that the target starts swapping if the memory is overcommitted.
>> it seems that the pages are dropped asynchronously.
>> 
>> Signed-off-by: Peter Lieven <pl@kamp.de>
>> ---
>> arch_init.c |   10 ++++++----
>> 1 file changed, 6 insertions(+), 4 deletions(-)
> 
>>             if (is_zero_page(p)) {
>>                 acct_info.dup_pages++;
>> -                bytes_sent = save_block_hdr(f, block, offset, cont,
>> -                                            RAM_SAVE_FLAG_COMPRESS);
>> -                qemu_put_byte(f, *p);
>> -                bytes_sent += 1;
>> +                if (!ram_bulk_stage) {
>> +                    bytes_sent = save_block_hdr(f, block, offset, cont,
>> +                                                RAM_SAVE_FLAG_COMPRESS);
>> +                    qemu_put_byte(f, 0);
>> +                }
>> +                bytes_sent++;
> 
> Logic is STILL wrong.  I pointed out in v2 that bytes_sent should not be
> incremented if you are not sending the page, so it needs to be inside
> the 'if (!ram_bulk_stage)'.

If its inside then bytes_sent will be -1 at the end if we skip a page. This would lead
to the raw page being sent. This way it is 0 what I think is correct.

> 
> Do we want to add a new migration statistic counter of how many zero
> pages we omitted sending during the bulk stage?

You mean sth like skipped zero pages?

I was also thinking of renaming dup_pages into zero_pages in the statistics,
but this could break someone relying on it so I left it as is.

Peter


> 
> -- 
> Eric Blake   eblake redhat com    +1-919-301-3266
> Libvirt virtualization library http://libvirt.org
>

Patch

diff --git a/arch_init.c b/arch_init.c
index 2b59454..c2cb40a 100644
--- a/arch_init.c
+++ b/arch_init.c
@@ -438,10 +438,12 @@  static int ram_save_block(QEMUFile *f, bool last_stage)
             bytes_sent = -1;
             if (is_zero_page(p)) {
                 acct_info.dup_pages++;
-                bytes_sent = save_block_hdr(f, block, offset, cont,
-                                            RAM_SAVE_FLAG_COMPRESS);
-                qemu_put_byte(f, *p);
-                bytes_sent += 1;
+                if (!ram_bulk_stage) {
+                    bytes_sent = save_block_hdr(f, block, offset, cont,
+                                                RAM_SAVE_FLAG_COMPRESS);
+                    qemu_put_byte(f, 0);
+                }
+                bytes_sent++;
             } else if (migrate_use_xbzrle()) {
                 current_addr = block->offset + offset;
                 bytes_sent = save_xbzrle_page(f, p, current_addr, block,