Re: [dc] draft-khasnabish-vmmi-problems-00.txt

[Vishwas Manral <[e-mail]>]

Hi Chris,

It is not about crashing alone, it could be a wrong entry added into
the table that could cause the packets to be routed wrongly. The idea
is about increasing the complexity in a single point of failure.

I guess this is not the most critical discussion and we can probably
move forward.

Thanks,
Vishwas


On 1/20/12, Christopher LILJENSTOLPE <[e-mail]> wrote:
> Greetings Vishwas,
>
> 	And I guess I am saying that I'm not sure I agree.  If a single process
> running can kill all the other processes on a given system (especially if it
> is in "user space" - which I would assume for control plane functions), I
> would say that there are more substantial issues with the architecture of
> that particular VM distribution.  Similar answer if someone actually put a
> control-plane engine (think routing protocols) in kernel space.  We, as a
> standards organization can't save the world from less-than-inteligent
> developers....
>
>
> 	Chris
>
> On 20Jan2012, at 15.47, Vishwas Manral wrote:
>
>> Hi Chistopher,
>>
>> I totally agree.
>>
>> The point I was making was that the hypervisor is the single point of
>> failure which can cause all the guest OS to fail. The more complex you
>> make the functionality the higer the chances of failure. So we should
>> work with that thought in mind.
>>
>> Thanks,
>> Vishwas
>>
>> On 1/20/12, Christopher LILJENSTOLPE <[e-mail]> wrote:
>>> Greetings,
>>>
>>> 	Another way of looking at it is a hypervisor is really an operating
>>> system
>>> distribution.  Many of the hypervisors out there have the network
>>> "switch"
>>> as a separate process (actually I believe all of them do).  So, if we are
>>> saying that networking intel doesn't belong in an OS distribution, that
>>> is a
>>> departure from current thinking :)
>>>
>>> 	Chris
>>>
>>> On 20Jan2012, at 15.31, Melinda Shore wrote:
>>>
>>>> On 01/20/2012 02:18 PM, Vishwas Manral wrote:
>>>>> An interesting thing to note is that the more the functionality you
>>>>> put in the hypervisor, the more you stress the single point of failure
>>>>> in the virtualized system.
>>>>
>>>> Well, there are a few ways to look at it.  For example,
>>>> the fewer components you've got the larger the mean time
>>>> between failures.  But aside from that it's been a
>>>> general rule of thumb that you want to minimize the
>>>> impact of failed components on non-failed components
>>>> (the fate sharing principle).
>>>>
>>>> At any rate the hypervisor (at least the ones with which
>>>> I'm familiar) basically *are* network devices - they
>>>> function as a switch, or even a NAT.  If you're going to
>>>> suggest that they can't be used to terminate control plane
>>>> sessions I hope there's a more compelling reason for it
>>>> than what has been offered so far.
>>>>
>>>> Melinda
>>>> _______________________________________________
>>>> dc mailing list
>>>> [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/dc
>>>
>>> --
>>> 李柯睿
>>> Check my PGP key here: https://www.asgaard.org/~cdl/cdl.asc
>>> Current vCard here: https://www.asgaard.org/~cdl/cdl.vcf
>>>
>>>
>
> --
> 李柯睿
> Check my PGP key here: https://www.asgaard.org/~cdl/cdl.asc
> Current vCard here: https://www.asgaard.org/~cdl/cdl.vcf
>
>