[Ocfs2-users] esx elevator=noop

[Brian Kroth]

At least in ESX you can setup various reservations and priority
weightings for various resources to unsure that some machines are
considered more important than others.  As to what actually happens in
practice, who knows.

Thanks for the info,
Brian

Herbert van den Bergh <herbert.van.den.bergh at oracle.com> 2010-01-15 11:20:
>
> I would assume that all bets are off if you're running a cluster inside  
> a vm.  There are no garantees for either I/O or cpu scheduling.
>
> Thanks,
> Herbert.
>
>
> On 01/15/2010 10:50 AM, Sunil Mushran wrote:
>> The deadline recommendation was for early el4 kernels that had a bug
>> in cfq. That bug was fixed years ago.
>>
>> I am unsure how using noop in guest will trigger starvation. Not that
>> I am recommending it. I have not thought about this much.
>>
>> On Jan 15, 2010, at 9:55 AM, Brian Kroth<bpkroth at gmail.com>  wrote:
>>
>>    
>>> http://lonesysadmin.net/2008/02/21/elevatornoop/
>>>
>>> I ran across this recently which describes, when operating in a
>>> virtual
>>> environment with shared storage, how to try and let the storage and
>>> hypervisor deal with arranging disk write operations in a more
>>> globally
>>> optimal way rather than having all the guests try to do it and muck it
>>> up.
>>>
>>> However, this is contrary to ocfs2 recommendation of using the
>>> deadline
>>> elevator.
>>>
>>> I'm just wondering if you have any comments one way or the other?
>>>
>>> My concern would be that while noop might make things globally optimal
>>> it would still allow starvation in a single guest which might lead to
>>> ocfs2 fencing.
>>>
>>> Thanks,
>>> Brian
>>>
>>> _______________________________________________
>>> Ocfs2-users mailing list
>>> Ocfs2-users at oss.oracle.com
>>> http://oss.oracle.com/mailman/listinfo/ocfs2-users
>>>      
>> _______________________________________________
>> Ocfs2-users mailing list
>> Ocfs2-users at oss.oracle.com
>> http://oss.oracle.com/mailman/listinfo/ocfs2-users
>>