[Ftk_hardware] crates

Paola Giannetti paola.giannetti at pi.infn.it
Thu Mar 24 06:11:28 CDT 2011


Yes, it requires a general re-thinking that is compatible with our 
Associative Memory schedule.
All of us are interested to ATCAs, this is why I was asking Ted to be 
the pioneer with the Data Formatter. It is also a demonstration the we 
listen the reviewers because we move (slowly) towards ATCAs.
The real point is that the Data Formatter is totally new, has no 
advantage from previous prototypes, has no particular timing constraints 
to be ready.
With Miercea we are introducing connectors and chips in a proto-AUX 
board that would allow to connect also the new AMBoard to  the EDRO 
before the end of data taking in 2012 (if everything goes ok with the 
new AMchip).
This means that the Data Formatter and the final AUX board can arrive 
later than 2012, but if the AMchip will work soon it will be a great 
advantage to see it in the vertical slice before the end of 2012.

I think this is enough interesting and still possible today, to decide 
to go ahead with VME at the maximum speed.
Il will allow us to build a very reliable system for 2015. The 2015 
system will be small, but will be able to take data immediately.

If really much more money and manpower will be available before 2015, a 
new design can start for the big production, but the money cannot come 
from Italy. We know  that we will not get an euro more than what INFN 
said (1,2 Meuros). We can save money from crates and racks if we use CDF 
stuff and put the saved money in more AMchip prototypes if they are 
necessary, we can swap items, but we cannot increase the budget.

If new collaborators will join with money they could build new LAMBs 
(smaller) with different AMchip packages, may be including serial 
transmission of data if they are really reach.

Regards
                            Paola

Jinlong Zhang wrote:
> Just add one thing: so far the ATCA crate has maximum 16 slots.
> This slot limit may cause more crates/racks and re-segmentation of the 
> data stream.
>
> On Thu, 24 Mar 2011, Paola Giannetti wrote:
>
>> I am sorry for the late answer, yesterday I couldn't see e-mails.
>>
>> I think we cannot use ATCA for the core crates for a lot of reasons that
>> I already explained, even at the review:
>> (1) we need early to test the system with the new chip, let's say next
>> year and it is not realistic to have a new AMBoard, new LAMB and new CPU
>> interface for the next year. It is very important to be able to measure
>> currents, capability to work at 100 MHz and with the new chip, before
>> the end of 2012. We will never do it if we change crate
>>
>> (2) We are ahead designing the LAMB, the AMBoard and the AMchip. The
>> suggestions of the reviewers are global, incoherent and do not take into
>> account the fact that we have already done a lot of work and the budget
>> will likely decrease not increase: (1) change package for the AMchip,
>> (2) go to 130 nm IBM instead of 65 nm TSMC, (3) use serial links in the
>> board instead of parallel buses, (4) get 30% more money (actually if we
>> go to serial links we need 400% more money)......  all this is totally
>> not realistic for our plan to be able to measure consumption, cooling
>> and performances in 2012. Before doing production, we need probably 2-3
>> AMChip prototypes, it requires time.
>>
>> (3) using VME we are compatible with the vertical slice, what we do for
>> the vertical slice is automatically good for the system. If we go to
>> ATCA all the work we do with the vertical slice is unuseful. We do not
>> have the manpower.
>>
>> (4) we already have 2 new VME crates, 2 new CPUs  in the lab, we cannot
>> put them in hold.
>>
>> (5) we are measuring the consumption and it seems to be less than
>> expected, may be a factor 2. This reduces a lot concerns  about 48 V (2
>> pins could be enough and separate GND return could be available).
>>
>> I really think ATCA for the core crate for 2015 is not feasible,
>> especially if in the same time we have to take data with the vertical
>> slice. Too much work in a too short time.
>> It is an interesting option for phase II. The LAMB will be redesigned
>> for that time, with new chips, new packages and new serial links.
>>
>> Regards
>>                            Paola
>>
>>
>>
>> Mel Shochet wrote:
>>>      There are now two suggestions for moving to ATCA.  Interboard data
>>> transfer in the Data Formatter crates could be made easier (the
>>> alternative being fibers between boards).  The Initial Design Review
>>> committee recommends that we consider ATCA for the core crates because
>>> of the committee's concern about power (our 48-volt solution) and
>>> cooling.  We will have to respond to this suggestion.  (A potential
>>> disadvantage is ATCA's 8U vs VME's 9U size).  Could we have an initial
>>> discussion of the issue at our meeting in 2 weeks (April 5)?  We could
>>> hear what the DF issues are from Fermilab and what the core crate 
>>> issues
>>> are from someone in Italy working on the AM and associated
>>> infrastructure.  Then we can decide on the next steps if any.
>>>
>>>                         Regards,
>>>                         Mel
>>>
>>>
>>>
>>> _______________________________________________
>>> Ftk_hardware mailing list
>>> Ftk_hardware at hep.uchicago.edu
>>> http://hep.uchicago.edu/mailman/listinfo/ftk_hardware
>>>
>>
>> _______________________________________________
>> Ftk_hardware mailing list
>> Ftk_hardware at hep.uchicago.edu
>> http://hep.uchicago.edu/mailman/listinfo/ftk_hardware
>>



More information about the Ftk_hardware mailing list