+1

Generally people have in mind the efficient use of resources when they
think of virtualization. Thus I would be in favor of sorting in
ascending order as that will approximate best fit and hence give
better resource utilization.

Mark

On Mon, Feb 18, 2013 at 10:55 AM, Aaron Coburn <[email protected]> wrote:
> +1
>
> I also like this idea. Making this a configurable option makes sense -- it 
> would be nice to have the default set to ascending order as you describe.
>
> Aaron C
>
>
> On Feb 18, 2013, at 10:42 AM, Aaron Peeler <[email protected]>
>  wrote:
>
>> I would like this ability, it would definitely simplify the mappings.
>>
>> I think it should be applied to both # or procs. and memory.
>>
>> I'm fine with either approach to just reserve it so it matches the min
>> valus or to have an option to choose which method.
>>
>> Aaron
>>
>> On Mon, Feb 18, 2013 at 10:31 AM, Josh Thompson <[email protected]> 
>> wrote:
>>> -----BEGIN PGP SIGNED MESSAGE-----
>>> Hash: SHA1
>>>
>>> Hi all,
>>>
>>> I'd like to get people's thoughts about reversing how computers are assigned
>>> for reservations relative to the specified amount of RAM for an image.
>>> Currently, the scheduler builds a list of computers that can fulfill a given
>>> reservation and orders them by specs of the machine by this priority:
>>>
>>> first by procspeed * proc number
>>> next by amount of RAM,
>>> finally by network speed
>>>
>>> Each of those is ordered in a descending order (i.e. best specs at the top).
>>> Then, the highest rated machine is given to the user.
>>>
>>> That algorithm came from our initial design back in 2004 when our nodes 
>>> didn't
>>> have lots of RAM, didn't have a high variability of contained RAM, didn't 
>>> have
>>> more RAM that some of the OSes could handle, and we weren't doing any
>>> virtualization.  The idea was that the user would get the best available
>>> machine for the reservation.
>>>
>>> Now, with nodes having very high amounts of RAM, a high variability of
>>> contained RAM, and having WinXP images that can't even use more than 4 GB of
>>> RAM, I'm wondering if ordering for RAM (and maybe all specs) should be
>>> reversed.  This would make it so that priority is given to assign a node 
>>> that
>>> just meets the specs for the image rather than assigning the best one
>>> available.  We're running in to cases where we have some bare metal nodes 
>>> with
>>> 24 GB or more of RAM, but still have WinXP images available to users.  We 
>>> have
>>> to map things so that the WinXP images cannot get deployed to the higher RAM
>>> nodes to keep from wasting the RAM when other users would like to have it.
>>> Things would be simplified if we could just have a more general pool and 
>>> have
>>> the scheduler take care of keeping resources from being wasted.
>>>
>>> What do others think?  I could also make it an option in conf.php as to 
>>> which
>>> method is used.  However, unless people feel it useful to keep the current
>>> method, it would be simpler to just reverse the ordering.  Also, if you 
>>> think
>>> it is a good idea to reverse it, should all specs be reversed, or just RAM?
>>>
>>> Thanks,
>>> Josh
>>> - --
>>> - -------------------------------
>>> Josh Thompson
>>> VCL Developer
>>> North Carolina State University
>>>
>>> my GPG/PGP key can be found at pgp.mit.edu
>>>
>>> All electronic mail messages in connection with State business which
>>> are sent to or received by this account are subject to the NC Public
>>> Records Law and may be disclosed to third parties.
>>> -----BEGIN PGP SIGNATURE-----
>>> Version: GnuPG v2.0.19 (GNU/Linux)
>>>
>>> iEYEARECAAYFAlEiSTwACgkQV/LQcNdtPQN3qQCeLaxbUg9Rh6F4mpQrcn1mh5jz
>>> VSEAn2C35CQCIqLnRUJFansQ5zKIhlNa
>>> =KVAL
>>> -----END PGP SIGNATURE-----
>>>
>>
>>
>>
>> --
>> Aaron Peeler
>> Program Manager
>> Virtual Computing Lab
>> NC State University
>>
>> All electronic mail messages in connection with State business which
>> are sent to or received by this account are subject to the NC Public
>> Records Law and may be disclosed to third parties.
>



-- 
Mark Gardner
--

Reply via email to