Dominique

1 Gb/se is enough, we try hard to design the system such that
network is not heavily used.

What kind of machines are you typically buying? (how many cores?)

1 disk per core (or 1 per 2 cores if you are getting machines
with many cores) would be useful.

As to disk size, depends on the pricing. I'd go with low-priced
disks, e.g. we care more about number of spindles than the actual
size (all new disks are now reasonably sized...).

Something like 8 cores, 8 x 1 or 2 TB disks, 16 GB RAM
would be a decent (and probably super very expensive)
start.

More on Thursday....

Jacek


On 09/24/2013 02:32 PM, Dominique Boutigny wrote:
> Hi Jacek,
>
> Running the current 50 nodes on the long term is not an option, they are
> really old now and are located in special racks (IBM iDataplex)
> requiring a dedicated cooling system which is also not maintained
> anymore. The idea would be to set up a qserv platform on more recent or
> even brand new hardware that would have more cores, memory and disk /
> server.
> I guess that it is important to have several distinct servers rather
> than just 3 or 4 servers with many core inside ? Should they be
> connected through 10 Gb/s links or is 1 Gb/s sufficient ? What is the
> ideal amount of disk space per server / per core ...
>
> Dominique
>
> Le 24/09/2013 14:20, Jacek Becla a écrit :
>> Dominique,
>>
>> We will discuss this Thursday and get back to you.
>> My feeling is that something like 8-10 nodes might
>> be the sweet spot if keeping 50 long term is not
>> an option. What is the annual cost to keep all 50
>> running? (assuming we just take out of service
>> machines that die).
>>
>> Jacek
>>
>>
>>
>>
>> On 09/23/2013 09:45 AM, Dominique Boutigny wrote:
>>> Hi,
>>>
>>> Let me add to Rachid's mail that it would be very useful to get a
>>> detailed description of both an ideal platform and a minimal one, I mean
>>> the minimum under which the platform is not useful anymore.
>>> Resources are limited and we will have to make choices.
>>>
>>> Best regards,
>>>
>>> Dominique
>>>
>>>
>>>
>>> Le 23/09/2013 03:03, Rachid Lemrani a écrit :
>>>> Dear all,
>>>>
>>>> The term "permanently" is not really correct. It's rather temporary
>>>> and short-term (to end of this year).
>>>> We are dealing here with old machines that are/were not maintained.
>>>> We expect a detailed description of the hardware needs for the qserv
>>>> test platform for 2014.
>>>> And our management/experts will evaluate the feasibility of the
>>>> request for the long-term.
>>>>
>>>> Best regards,
>>>> Rachid
>>>>
>>>>
>>>> On 23 sept. 2013, at 10:06, Fabrice Jammes <[log in to unmask]>
>>>> wrote:
>>>>
>>>>> Hello Daniel, Douglas, Jacek, and Serge,
>>>>>
>>>>> Good news ! CC-IN2P3 agree to keep 50 machines running permanently
>>>>> for Qserv development purpose.
>>>>> Here's two questions from Rachid Lemrani about these remaining
>>>>> machines at CC-IN2P3 :
>>>>>
>>>>> - Do you havea preference on the set of machines to keep (in ccqserv
>>>>> set {001 .. 320}), or can the CC-IN2P3 engineers choose it
>>>>> themselves(depending on racks for example)?
>>>>> - Do you still need the"first" installed qserv machines (ccqserv {01
>>>>> .. 15})?Can it be stopped ?
>>>>>
>>>>> Thanks,
>>>>>
>>>>> Fabrice
>>>
>>
>

########################################################################
Use REPLY-ALL to reply to list

To unsubscribe from the QSERV-L list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=QSERV-L&A=1