How to calculate the crashing time cost in live migration

How to calculate the crashing time cost in live migration

[Yunbin Wang]

[-- Attachment #1.1: Type: text/plain, Size: 349 bytes --]

Hi all,
I'm try to improve performance of live migration of Xen, and I implemented
a new method to migrate, so I want to test whether the new method can bring
a better performance.

I test the method from two aspects, the migration time and crash time. The
migration time is easy to get, but how can I calculate the crash time?

Thanks,
Yunbin Wang

[-- Attachment #1.2: Type: text/html, Size: 443 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: How to calculate the crashing time cost in live migration

[Andrew Cooper]

[-- Attachment #1.1: Type: text/plain, Size: 531 bytes --]

On 22/12/2013 13:23, Yunbin Wang wrote:
> Hi all,
> I'm try to improve performance of live migration of Xen, and I
> implemented a new method to migrate, so I want to test whether the new
> method can bring a better performance.
>
> I test the method from two aspects, the migration time and crash time.
> The migration time is easy to get, but how can I calculate the crash time?
>
> Thanks,
> Yunbin Wang

Can you explain what you mean by the "crash time", and perhaps also
explain how you are measuring migration time.

~Andrew

[-- Attachment #1.2: Type: text/html, Size: 1198 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: How to calculate the crashing time cost in live migration

[Yunbin Wang]

[-- Attachment #1.1: Type: text/plain, Size: 1194 bytes --]

Hi Cooper,
Thanks for your reply.

Firstly, I explain how to measure migration time. When live migration
needed, we should enter the following command
xm migrate --live <domain> <target>, so I just measure the cost time by
this command as migration time.

Secondly, I explain the meaning of "crash time" aforementioned. This "crash
time", I mean down time. Although using live migration technology, VM can
be still in service status during migration, but there's still a small
period, the VM will be out of service. So I want to measure the period VM
out of service.

Thanks,
Yunbin Wang


2013/12/22 Andrew Cooper <andrew.cooper3@citrix.com>

>  On 22/12/2013 13:23, Yunbin Wang wrote:
>
> Hi all,
> I'm try to improve performance of live migration of Xen, and I implemented
> a new method to migrate, so I want to test whether the new method can bring
> a better performance.
>
>  I test the method from two aspects, the migration time and crash time.
> The migration time is easy to get, but how can I calculate the crash time?
>
>  Thanks,
> Yunbin Wang
>
>
> Can you explain what you mean by the "crash time", and perhaps also
> explain how you are measuring migration time.
>
> ~Andrew
>

[-- Attachment #1.2: Type: text/html, Size: 2130 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: How to calculate the crashing time cost in live migration

[Kai Huang]

[-- Attachment #1.1: Type: text/plain, Size: 2052 bytes --]

If I understand correctly, from hypervisor's view, there are two situations
that VM will be out of service. The first one is when userland calls
hypercall to get dirty pages, during which the domain will be paused. And
the second one is at last stage of migration, the domain will be suspended
and transmitted to remote. For the first case, we can simply add
__trace_var to the beginning and end of domain_log_dirty_op, and use
xentrace to get the time. For the second case, I am not familiar with it,
but if it's also done in one hypercall, it can be done as the first case.

-Kai


On Mon, Dec 23, 2013 at 2:20 PM, Yunbin Wang <wangyunbin1989@gmail.com>wrote:

> Hi Cooper,
> Thanks for your reply.
>
> Firstly, I explain how to measure migration time. When live migration
> needed, we should enter the following command
> xm migrate --live <domain> <target>, so I just measure the cost time by
> this command as migration time.
>
> Secondly, I explain the meaning of "crash time" aforementioned. This
> "crash time", I mean down time. Although using live migration technology,
> VM can be still in service status during migration, but there's still a
> small period, the VM will be out of service. So I want to measure the
> period VM out of service.
>
> Thanks,
> Yunbin Wang
>
>
> 2013/12/22 Andrew Cooper <andrew.cooper3@citrix.com>
>
>>  On 22/12/2013 13:23, Yunbin Wang wrote:
>>
>> Hi all,
>> I'm try to improve performance of live migration of Xen, and I
>> implemented a new method to migrate, so I want to test whether the new
>> method can bring a better performance.
>>
>>  I test the method from two aspects, the migration time and crash time.
>> The migration time is easy to get, but how can I calculate the crash time?
>>
>>  Thanks,
>> Yunbin Wang
>>
>>
>> Can you explain what you mean by the "crash time", and perhaps also
>> explain how you are measuring migration time.
>>
>> ~Andrew
>>
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
>
>

[-- Attachment #1.2: Type: text/html, Size: 3420 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel