xen and virtual iron

xen and virtual iron

[Sean Harper]

At Linuxworld a company called Virtual Iron is demoing a product that,
like Xen or VMWare, can break physical machines into virtual machines.

However, unlike Xen or VMWare, this product can also aggregate physical
machines into virtual machines.  In other words, it is possible for the
user to specify that 3.5 cpus from 2 machines (2 from 1 and 1.5 from
another) be assigned to a virtual machine. When linux boots on that
virtual machine it simply looks like a 4 cpu machine (but 1 of the cpus
is slower). Presumably there is a pretty big performance penalty for
sharing across machines, which they mitigate to some extent by requiring
Infiniband.

It seems like most of the tricky work is around caching to optimize the
performance across the slower communication bus (when sharing between
machines).

Is this sort of capability something that anyone involved with Xen is
thinking about? Despite the performance impacts I can imagine this being
a useful tool in big datacenters.



-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_ide95&alloc_id\x14396&op=click

Re: xen and virtual iron

[Andrew Theurer]

Sean Harper wrote:

>At Linuxworld a company called Virtual Iron is demoing a product that,
>like Xen or VMWare, can break physical machines into virtual machines.
>
>However, unlike Xen or VMWare, this product can also aggregate physical
>machines into virtual machines.  In other words, it is possible for the
>user to specify that 3.5 cpus from 2 machines (2 from 1 and 1.5 from
>another) be assigned to a virtual machine. When linux boots on that
>virtual machine it simply looks like a 4 cpu machine (but 1 of the cpus
>is slower). Presumably there is a pretty big performance penalty for
>sharing across machines, which they mitigate to some extent by requiring
>Infiniband.
>
>It seems like most of the tricky work is around caching to optimize the
>performance across the slower communication bus (when sharing between
>machines).
>  
>
Hmm, sounds like a NUMA system, with a relatively low throughput and 
high latency interconnect.   Not sure if anyone would want to do this on 
a workload where performance matters.

I suppose with a layer of abstraction like Xen, doing something like 
this is feasible.  You could leverage the NUMA code in the linux kernel, 
but I would think you would need a very highly parallel workload to make 
this effective, and if you have that, a cluster setup would probably 
work just as well anyway.

-Andrew Theurer


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click

Re: xen and virtual iron

[Moshe Bar]

Virtual Iron seems to me to be a mix of Xen and openMosix, combined 
with a Distributed Shared Memory (which openMosix has as an option).

Scaling up applications in a DSM environment is hugely dependent on the 
application. If all you have are HPC type applications, then Virtual 
Iron might scale nicely, but then why would you pay for something which 
stuff like openMosix provides for free? For most business apps, I 
suspect that a DSM-based solution will not scale at all, or have a 
detrimental impact on performance/thruput, due to the issues with true 
and false sharing.

They do have a pretty console application, though.

Moshe





On Feb 16, 2005, at 7:57 AM, Andrew Theurer wrote:

> Sean Harper wrote:
>
>> At Linuxworld a company called Virtual Iron is demoing a product that,
>> like Xen or VMWare, can break physical machines into virtual machines.
>>
>> However, unlike Xen or VMWare, this product can also aggregate 
>> physical
>> machines into virtual machines.  In other words, it is possible for 
>> the
>> user to specify that 3.5 cpus from 2 machines (2 from 1 and 1.5 from
>> another) be assigned to a virtual machine. When linux boots on that
>> virtual machine it simply looks like a 4 cpu machine (but 1 of the 
>> cpus
>> is slower). Presumably there is a pretty big performance penalty for
>> sharing across machines, which they mitigate to some extent by 
>> requiring
>> Infiniband.
>>
>> It seems like most of the tricky work is around caching to optimize 
>> the
>> performance across the slower communication bus (when sharing between
>> machines).
>>
> Hmm, sounds like a NUMA system, with a relatively low throughput and 
> high latency interconnect.   Not sure if anyone would want to do this 
> on a workload where performance matters.
>
> I suppose with a layer of abstraction like Xen, doing something like 
> this is feasible.  You could leverage the NUMA code in the linux 
> kernel, but I would think you would need a very highly parallel 
> workload to make this effective, and if you have that, a cluster setup 
> would probably work just as well anyway.
>
> -Andrew Theurer
>
>
> -------------------------------------------------------
> SF email is sponsored by - The IT Product Guide
> Read honest & candid reviews on hundreds of IT Products from real 
> users.
> Discover which products truly live up to the hype. Start reading now.
> http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/xen-devel
>



-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click