Status of Fault Tolerance feature?

Status of Fault Tolerance feature?

[Andres Toomsalu]

Hi,

Could anyone shed a light what happened to Kemari project and are there any upcoming development planned in order to provide continous non-blocking VM checkpointing and VM HA with state replication?

Kind regards,
-- 
----------------------------------------------
Andres Toomsalu, andres@opennodecloud.com
http://www.opennodecloud.com

Re: Status of Fault Tolerance feature?

[Brian Jackson]

On Mon, 21 Jan 2013 14:24:12 +0200
Andres Toomsalu <andres@opennodecloud.com> wrote:

> Hi,
> 
> Could anyone shed a light what happened to Kemari project and are
> there any upcoming development planned in order to provide continous
> non-blocking VM checkpointing and VM HA with state replication?
> 
> Kind regards,

The project hasn't been actively developed in years and there has been
no public information about it in probably longer. So state is
"unknown".

Re: Status of Fault Tolerance feature?

[Andres Toomsalu]

But is there any other projects in (planned) development with the same goal(s)?
Im just really puzzled that while QEMU/KVM being kind a mature solution already no true fault tolerance/HA solutions exist (Im aware about stateless HA solutions with RHCS etc stacks -  but its hardly the "true" HA) - and if I get it correctly - no real plans/development in that direction also near-term?

Kind regards,
-- 
----------------------------------------------
Andres Toomsalu, andres@opennodecloud.com
http://www.opennodecloud.com

On 29.01.2013, at 0:46, Brian Jackson wrote:

> On Mon, 21 Jan 2013 14:24:12 +0200
> Andres Toomsalu <andres@opennodecloud.com> wrote:
> 
>> Hi,
>> 
>> Could anyone shed a light what happened to Kemari project and are
>> there any upcoming development planned in order to provide continous
>> non-blocking VM checkpointing and VM HA with state replication?
>> 
>> Kind regards,
> 
> The project hasn't been actively developed in years and there has been
> no public information about it in probably longer. So state is
> "unknown".

Re: Status of Fault Tolerance feature?

[Brian Jackson]

On Tue, 29 Jan 2013 15:16:13 +0200
Andres Toomsalu <andres@opennodecloud.com> wrote:

> But is there any other projects in (planned) development with the
> same goal(s)?


I haven't heard of any. But then again, a lot of things get developed
in secret and then dumped on the community.


> Im just really puzzled that while QEMU/KVM being kind a
> mature solution already no true fault tolerance/HA solutions exist
> (Im aware about stateless HA solutions with RHCS etc stacks -  but
> its hardly the "true" HA) - and if I get it correctly - no real
> plans/development in that direction also near-term?


Most people that I know that have tried similar solutions on other
products give up on it because the performance is abysmal. It's
generally faster and better tested to do this stuff at the application
layer.


> 
> Kind regards,

Re: Status of Fault Tolerance feature?

[Andres Toomsalu]

On 29.01.2013, at 17:20, Brian Jackson wrote:

> On Tue, 29 Jan 2013 15:16:13 +0200
> Andres Toomsalu <andres@opennodecloud.com> wrote:
> 
>> But is there any other projects in (planned) development with the
>> same goal(s)?
> 
> 
> I haven't heard of any. But then again, a lot of things get developed
> in secret and then dumped on the community.

Sure.

> 
> 
>> Im just really puzzled that while QEMU/KVM being kind a
>> mature solution already no true fault tolerance/HA solutions exist
>> (Im aware about stateless HA solutions with RHCS etc stacks -  but
>> its hardly the "true" HA) - and if I get it correctly - no real
>> plans/development in that direction also near-term?
> 
> 
> Most people that I know that have tried similar solutions on other
> products give up on it because the performance is abysmal. It's
> generally faster and better tested to do this stuff at the application
> layer.

I've been looking into (somewhat) hypervisor agnostic solutions - eg general linux checkpoint-restore solutions - like DMTCP (dmtcp.sf.net) and CRIU (criu.org).
There is actually proof-of-concept DMTCP implementation for KVM - described in this paper: http://arxiv.org/pdf/1212.1787v1.pdf
CRIU currently supports only linux containers (OpenVZ, LXC) - but probaly it would be possible to add support also for QEMU/KVM by similar approach as DMTCP KVM plugin does it.

If I understand correctly DMTCP approach is a kind of non-blocking solution with possibly acceptable perfomance tradeoff.
I would see great benefit of having kind of a "generic" checkpoint/restore mechanism with support for multiple hypervisors - which could be basis also for a future fault tolerance/HA solutions.  

> 
> 
>> 
>> Kind regards,
>