[Pafgbt] PAF beamformer size and cost

[Rick Fisher]

Any help I can get on sketching out a sensible computing architecture will 
be muh appreciated.  Paul and I kicked around a few issues in the coffee 
room this morning, but I'll need lots of tutoring.

For hardware distribution, it seems to make sense to keep the ROACH boards 
and ethernet switch in the same room as the data storage computer.  That's 
another argument for putting the first beamformer in the lab.

What issues do we need to address?  M&C interface.  Data pipelining.  Data 
archiving.  ? ? ?  Is the GUPPI system a good starting point?

Rick

On Mon, 8 Feb 2010, Scott Ransom wrote:

> On Monday 08 February 2010 10:09:59 am John Ford wrote:
>>> Roughly 50 MB/s is probably a good stake in the ground for the
>>> limits of output data bandwidth and storage limitations.  We can
>>> probably plan on these capabilities doubling every 2 years or so,
>>> but it's not a smooth function of time, of course.
>>
>> It's actually the data management that's a problem, not the
>>  collection speeds.  It's very easy to do 50-100 MB/sec data rates to
>>  disk over 10 gigabit Ethernet.  40 gigabit infiniband is available
>>  and almost as cheap as 10 gbe right now.  It's what to do with all
>>  those terabytes once you have it on disk.
>>
>> I think that any of these designs should include the design of a
>>  computer system (software and hardware) to complete the entire
>>  scientific data flow, and not just the data acquisition.  We always
>>  think about it a little, and then punt it to the future, and we're
>>  always sorry...  (I think Scott will agree!)
>
> Oh I definitely agree.  It is the same with active RFI excision...
>
> Scott
>
>
>