N 2 GC for VLBI S.Pogrebenko, JIVE, Dwingeloo, NL, ( [email protected] )[email protected] 29.06.2006, Next Generation Correlators Workshop, Groningen,

NN22GC for VLBIGC for VLBIS.Pogrebenko, JIVE, Dwingeloo, NL, ( [email protected] )

29.06.2006 , Next Generation Correlators Workshop, Groningen, NL

Did I say N2GC?..

VLBI NGC is already there,

Just take EVLA, it can do Mk5global VLBI as a background task,global Mk6 will take just a fractionof its resources

And we can consider something a little bit more funny

Like this:Like this:4-16 stations,4-16 stations,

1-4 Gbps/station1-4 Gbps/station

VLBI solutionVLBI solution(space craft tracking enabled)(space craft tracking enabled)

or or baby-SKA beam formerbaby-SKA beam former

or or FPA descramblerFPA descrambler

((for just under 1 kW for just under 1 kW

thermalthermal)) COMMUNICATION LOGIC

CELLCELLCELLCELL

CELLCELLCELLCELL

CELLCELLCELLCELL

CELLCELLCELLCELL

COMMUNICATION LOGIC

Single board, 16 CELLs,16 of 10G fiber connections

S. Pogrebenko, 10.08.2005 , 04.10.2005 & 12.10.2005, 29.06.2006

It’ll just take about 1 Liter/second water flow to absorb 30 kW of It’ll just take about 1 Liter/second water flow to absorb 30 kW of thermal power to get some 1-2 Tbps aggregated 16-32 station thermal power to get some 1-2 Tbps aggregated 16-32 station

solutionsolution

S. Pogrebenko, 10.08.2005 , 04.10.2005 & 12.10.2005, 29.06.2006

Using “almost” off the shelf Supercomputer-on-chip solution (IBM CELL is an example),

In 5 years of work 2xEVLAs can be squeezed into under 30 kW thermal envelope

for about the same design and manufacturing cost as EVLA-1

Future is bright!Future is bright!Only Sky is a limit!Only Sky is a limit!

Image Credit: NASA, ESA, B. Mobasher (Space Telescope Science Institute and the European Space Agency)

Only SKA is a limit!Only SKA is a limit!

Oops… Did I say Sky?Oops… Did I say Sky?

There are two thick brick wallson the way to wide field

VLBI+SKAFirst one is Nstations square,

So thick that even Super-FXcan’t break it

Second one is even more thicker: When the output data rate exceeds

the input one, the Correlator becomes just a waste of CPU power, money and senseand don’t worth a real time of our real life

Solution?

Shift a gearfrom

van Sittert and Zernikeparadigm to Kirchghoff

S. Pogrebenko, 30.01.1996, 10.08.2005 , 04.10.2005, 12.10.2005, 29.06.2006

And to make it really interesting,And to make it really interesting,let’s talk about scalable let’s talk about scalable

NNssLogLog22(N(Nss) paradigm) paradigm

S. Pogrebenko, 30.01.1996, 10.08.2005 , 04.10.2005, 12.10.2005, 29.06.2006

What I like about software What I like about software correlators on advanced HW correlators on advanced HW

platforms is an easy shift platforms is an easy shift between processing paradigmsbetween processing paradigms

What I like even more is that it

allows to concentrate on the essence of

VLBI,Adding more refined functionality with no

other costs but processing time

And that allows to do a lot of fancy things