[Qemu-devel] [PATCH] migration: Increase default max_downtime from 30ms to 300ms

[Qemu-devel] [PATCH] migration: Increase default max_downtime from 30ms to 300ms

[Alexey Kardashevskiy]

The existing timeout is 30ms which on 100MB/s (1Gbit) gives us
3MB/s rate maximum. If we put some load on the guest, it is easy to
get page dirtying rate too big so live migration will never complete.
In the case of libvirt that means that the guest will be stopped
anyway after a timeout specified in the "virsh migrate" command and
this normally generates even bigger delay.

This changes max_downtime to 300ms which seems to be more
reasonable value.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 migration.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/migration.c b/migration.c
index e0e24d4..02bbce9 100644
--- a/migration.c
+++ b/migration.c
@@ -144,7 +144,7 @@ void process_incoming_migration(QEMUFile *f)
  * the choice of nanoseconds is because it is the maximum resolution that
  * get_clock() can achieve. It is an internal measure. All user-visible
  * units must be in seconds */
-static uint64_t max_downtime = 30000000;
+static uint64_t max_downtime = 300000000;
 
 uint64_t migrate_max_downtime(void)
 {
-- 
1.8.4.rc4

Re: [Qemu-devel] [PATCH] migration: Increase default max_downtime from 30ms to 300ms

[Markus Armbruster]

Adding Juan.

Alexey Kardashevskiy <aik@ozlabs.ru> writes:

> The existing timeout is 30ms which on 100MB/s (1Gbit) gives us
> 3MB/s rate maximum. If we put some load on the guest, it is easy to
> get page dirtying rate too big so live migration will never complete.
> In the case of libvirt that means that the guest will be stopped
> anyway after a timeout specified in the "virsh migrate" command and
> this normally generates even bigger delay.
>
> This changes max_downtime to 300ms which seems to be more
> reasonable value.
>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  migration.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/migration.c b/migration.c
> index e0e24d4..02bbce9 100644
> --- a/migration.c
> +++ b/migration.c
> @@ -144,7 +144,7 @@ void process_incoming_migration(QEMUFile *f)
>   * the choice of nanoseconds is because it is the maximum resolution that
>   * get_clock() can achieve. It is an internal measure. All user-visible
>   * units must be in seconds */
> -static uint64_t max_downtime = 30000000;
> +static uint64_t max_downtime = 300000000;
>  
>  uint64_t migrate_max_downtime(void)
>  {

Re: [Qemu-devel] [PATCH] migration: Increase default max_downtime from 30ms to 300ms

[Alexey Kardashevskiy]

On 04/30/2014 11:10 AM, Alexey Kardashevskiy wrote:
> On 04/13/2014 12:38 AM, Alexey Kardashevskiy wrote:
>> On 03/27/2014 08:01 PM, Markus Armbruster wrote:
>>> Adding Juan.
>>
>>
>> Ping?
> 
> 
> Ping?
> 
> 

Ping?


>>
>>>
>>> Alexey Kardashevskiy <aik@ozlabs.ru> writes:
>>>
>>>> The existing timeout is 30ms which on 100MB/s (1Gbit) gives us
>>>> 3MB/s rate maximum. If we put some load on the guest, it is easy to
>>>> get page dirtying rate too big so live migration will never complete.
>>>> In the case of libvirt that means that the guest will be stopped
>>>> anyway after a timeout specified in the "virsh migrate" command and
>>>> this normally generates even bigger delay.
>>>>
>>>> This changes max_downtime to 300ms which seems to be more
>>>> reasonable value.
>>>>
>>>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>>>> ---
>>>>  migration.c | 2 +-
>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/migration.c b/migration.c
>>>> index e0e24d4..02bbce9 100644
>>>> --- a/migration.c
>>>> +++ b/migration.c
>>>> @@ -144,7 +144,7 @@ void process_incoming_migration(QEMUFile *f)
>>>>   * the choice of nanoseconds is because it is the maximum resolution that
>>>>   * get_clock() can achieve. It is an internal measure. All user-visible
>>>>   * units must be in seconds */
>>>> -static uint64_t max_downtime = 30000000;
>>>> +static uint64_t max_downtime = 300000000;
>>>>  
>>>>  uint64_t migrate_max_downtime(void)
>>>>  {
>>
>>
> 
> 


-- 
Alexey

Re: [Qemu-devel] [PATCH] migration: Increase default max_downtime from 30ms to 300ms

[Juan Quintela]

Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
> On 04/13/2014 12:38 AM, Alexey Kardashevskiy wrote:
>> On 03/27/2014 08:01 PM, Markus Armbruster wrote:
>>> Adding Juan.
>> 
>> 
>> Ping?

Patch is OK for me.

As sender says, with guests doing anything/bigger than 1GB RAM is
basically implosible to get into the 30ms downtime.

Later, Juan.

>
>
> Ping?
>
>
>> 
>>>
>>> Alexey Kardashevskiy <aik@ozlabs.ru> writes:
>>>
>>>> The existing timeout is 30ms which on 100MB/s (1Gbit) gives us
>>>> 3MB/s rate maximum. If we put some load on the guest, it is easy to
>>>> get page dirtying rate too big so live migration will never complete.
>>>> In the case of libvirt that means that the guest will be stopped
>>>> anyway after a timeout specified in the "virsh migrate" command and
>>>> this normally generates even bigger delay.
>>>>
>>>> This changes max_downtime to 300ms which seems to be more
>>>> reasonable value.
>>>>
>>>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>>>> ---
>>>>  migration.c | 2 +-
>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/migration.c b/migration.c
>>>> index e0e24d4..02bbce9 100644
>>>> --- a/migration.c
>>>> +++ b/migration.c
>>>> @@ -144,7 +144,7 @@ void process_incoming_migration(QEMUFile *f)
>>>>   * the choice of nanoseconds is because it is the maximum resolution that
>>>>   * get_clock() can achieve. It is an internal measure. All user-visible
>>>>   * units must be in seconds */
>>>> -static uint64_t max_downtime = 30000000;
>>>> +static uint64_t max_downtime = 300000000;
>>>>  
>>>>  uint64_t migrate_max_downtime(void)
>>>>  {
>> 
>> 

Re: [Qemu-devel] [PATCH] migration: Increase default max_downtime from 30ms to 300ms

[Alexey Kardashevskiy]

On 05/05/2014 09:20 PM, Juan Quintela wrote:
> Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
>> On 04/13/2014 12:38 AM, Alexey Kardashevskiy wrote:
>>> On 03/27/2014 08:01 PM, Markus Armbruster wrote:
>>>> Adding Juan.
>>>
>>>
>>> Ping?
> 
> Patch is OK for me.



Who else needs to be ok to get this in upstream? :) Thanks!



> As sender says, with guests doing anything/bigger than 1GB RAM is
> basically implosible to get into the 30ms downtime.
> 
> Later, Juan.
> 
>>
>>
>> Ping?
>>
>>
>>>
>>>>
>>>> Alexey Kardashevskiy <aik@ozlabs.ru> writes:
>>>>
>>>>> The existing timeout is 30ms which on 100MB/s (1Gbit) gives us
>>>>> 3MB/s rate maximum. If we put some load on the guest, it is easy to
>>>>> get page dirtying rate too big so live migration will never complete.
>>>>> In the case of libvirt that means that the guest will be stopped
>>>>> anyway after a timeout specified in the "virsh migrate" command and
>>>>> this normally generates even bigger delay.
>>>>>
>>>>> This changes max_downtime to 300ms which seems to be more
>>>>> reasonable value.
>>>>>
>>>>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>>>>> ---
>>>>>  migration.c | 2 +-
>>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/migration.c b/migration.c
>>>>> index e0e24d4..02bbce9 100644
>>>>> --- a/migration.c
>>>>> +++ b/migration.c
>>>>> @@ -144,7 +144,7 @@ void process_incoming_migration(QEMUFile *f)
>>>>>   * the choice of nanoseconds is because it is the maximum resolution that
>>>>>   * get_clock() can achieve. It is an internal measure. All user-visible
>>>>>   * units must be in seconds */
>>>>> -static uint64_t max_downtime = 30000000;
>>>>> +static uint64_t max_downtime = 300000000;
>>>>>  
>>>>>  uint64_t migrate_max_downtime(void)
>>>>>  {
>>>
>>>


-- 
Alexey

Re: [Qemu-devel] [PATCH] migration: Increase default max_downtime from 30ms to 300ms

[Juan Quintela]

Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
> On 05/05/2014 09:20 PM, Juan Quintela wrote:
>> Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
>>> On 04/13/2014 12:38 AM, Alexey Kardashevskiy wrote:
>>>> On 03/27/2014 08:01 PM, Markus Armbruster wrote:
>>>>> Adding Juan.
>>>>
>>>>
>>>> Ping?
>> 
>> Patch is OK for me.
>
>
>
> Who else needs to be ok to get this in upstream? :) Thanks!

Peter?
David?
Paolo? (I think to remember that they are ok with it)

Thanks, Juan.

Re: [Qemu-devel] [PATCH] migration: Increase default max_downtime from 30ms to 300ms

[Paolo Bonzini]

Il 04/06/2014 12:25, Juan Quintela ha scritto:
>> >
>> > Who else needs to be ok to get this in upstream? :) Thanks!
> Peter?
> David?
> Paolo? (I think to remember that they are ok with it)

Certainly okay by me.

Paolo

Re: [Qemu-devel] [PATCH] migration: Increase default max_downtime from 30ms to 300ms

[Amit Shah]

On (Thu) 27 Mar 2014 [14:57:26], Alexey Kardashevskiy wrote:
> The existing timeout is 30ms which on 100MB/s (1Gbit) gives us
> 3MB/s rate maximum. If we put some load on the guest, it is easy to
> get page dirtying rate too big so live migration will never complete.
> In the case of libvirt that means that the guest will be stopped
> anyway after a timeout specified in the "virsh migrate" command and
> this normally generates even bigger delay.
> 
> This changes max_downtime to 300ms which seems to be more
> reasonable value.
> 
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>

Reviewed-by: Amit Shah <amit.shah@redhat.com>

		Amit

Re: [Qemu-devel] [PATCH] migration: Increase default max_downtime from 30ms to 300ms

[Alexey Kardashevskiy]

On 06/06/2014 10:50 PM, Amit Shah wrote:
> On (Thu) 27 Mar 2014 [14:57:26], Alexey Kardashevskiy wrote:
>> The existing timeout is 30ms which on 100MB/s (1Gbit) gives us
>> 3MB/s rate maximum. If we put some load on the guest, it is easy to
>> get page dirtying rate too big so live migration will never complete.
>> In the case of libvirt that means that the guest will be stopped
>> anyway after a timeout specified in the "virsh migrate" command and
>> this normally generates even bigger delay.
>>
>> This changes max_downtime to 300ms which seems to be more
>> reasonable value.
>>
>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> 
> Reviewed-by: Amit Shah <amit.shah@redhat.com>

Cool, thanks!

But what is next? The patch is > 2 months old and it is simple and
everything but it did not make it to any tree so far :)


-- 
Alexey

Re: [Qemu-devel] [PATCH] migration: Increase default max_downtime from 30ms to 300ms

[Amit Shah]

On (Sat) 07 Jun 2014 [02:48:14], Alexey Kardashevskiy wrote:
> On 06/06/2014 10:50 PM, Amit Shah wrote:
> > On (Thu) 27 Mar 2014 [14:57:26], Alexey Kardashevskiy wrote:
> >> The existing timeout is 30ms which on 100MB/s (1Gbit) gives us
> >> 3MB/s rate maximum. If we put some load on the guest, it is easy to
> >> get page dirtying rate too big so live migration will never complete.
> >> In the case of libvirt that means that the guest will be stopped
> >> anyway after a timeout specified in the "virsh migrate" command and
> >> this normally generates even bigger delay.
> >>
> >> This changes max_downtime to 300ms which seems to be more
> >> reasonable value.
> >>
> >> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> > 
> > Reviewed-by: Amit Shah <amit.shah@redhat.com>
> 
> Cool, thanks!
> 
> But what is next? The patch is > 2 months old and it is simple and
> everything but it did not make it to any tree so far :)

I'll pester Juan to get it in :-)

		Amit