DRAFT
Send comments to Eric
Summary
of the G3 and Myrinet programming model
19th October
2001
Scribe: Eric Cano
Agenda
- Layout
of the software for the DAQ (including low level access)
- Programming models for G3, Myrinet and VME
Present
Eric Cano, Sergio Cittolin, Dominique Gigi, Johannes
Gutleber, Frans Meijers, Luciano Orsini, Christoph Schwick
Layout of the software of the DAQ
After several discussion a picture of the layout
of the DAQ software came out. The idea of a software bundle (which we would
call XDAQ) was brought out.
Sergio remided that up to end 2002, we were still in
R&D time, which means our tasks are :
Programming model for the G3 (FED receiver)
Several programming models for the G3 we proposed, but all rely on FIFOs used
to send the G3 free buffer lists. G3 sends data to the buffer when it receives
it and send back the buffers list and event size (an event can span on several
buffers) to a FIFO in the host. After discussion with Frans it turned out that
a similar model can be used to send date directly from a G3 to the myrinet board
(FRL-like scheme).
The problem of endian swapping of data on the G3 was raised, but some said
we (DAQ) don't hav to access the payload of the events, only headers. So we
might swap by software the few header registers we're intersted in. (this was
undecided).
We made a list of all the cases in which we want to use the boards.
- FED development KIT : G3 as an slink receiver to host memory.
- Myrinet as a receiver : Myrinet to host memory.
- FRL-like scheme : G3 to Myrinet board
- Local DAQ : VME to host memory (purely software)
In the latter scheme, we want to rely as much as possible on drivers provided
by the VME board constructor. This latter scheme is also FED dependant (local
DAQ) so the DAQ group con only provide guideline and toolkits.