Pulsar_Machine_Development_Status_Updates_from_October_2007 - david-macmahon/wiki_convert_test GitHub Wiki

In October 2007 there was a round of updates on the status of pulsar machines being developed using CASPER hardware and tools.

Request for Status Updates

We've been in touch with most of you individually regarding CASPER pulsar hardware developments. As the Berkeley workshop was three months ago by now, we want to make sure everyone is up to date with developments elsewhere. We would therefore like to invite you to group-reply with a short summary of the current status of your CASPER-related pulsar hardware plans or development. You can also reply to just this group alias:

casper-pulsars at domain astro[dot]berkeley[dot]edu

Thanks. Kind regards,

Joeri van Leeuwen

Glen Jones, Caltech

Hello from Caltech,

We finally received our first iBOB last week, and two days ago I got the Pocket Spectrometer code up and running and took my first spectra. We are expecting to receive another iBOB and a BEE2 in the next week or so. In the mean time, I have familiarized myself with basically all of the "Green" blocks in the CASPER library, which took longer than I'd originally expected. I found the documentation to still be lacking. I think some of the best documentation is working examples showing how to use each block, which is basically what I did to learn how they all work. I plan to polish this up and distribute it so that others can benefit from these examples.

My main priority now is getting a spectrometer working at 2 Gsps, which is also proving more difficult than I'd originally expected. When the BEE2 arrives, I want to be able to digitize an I and Q signal at 1 GHz each for a total of 2 GHz of bandwidth, and then form a spectrum and hopefully also cross multiply it with another similar 2 GHz signal to form a full Stokes spectrometer. On the pulsar front, I have made a new version of the Vector Accumulator block which I think is easier to understand, and more importantly, lends itself readily to applying incoherent dedispersion on chip. I have worked out how to wire up the FPGA incoherent dedispersion, but have not finished implementing it because the basic spectroscopy is a more pressing issue.

I am looking to purchase a 10 GbE to 1 GbE switch soon to get started testing how well the data streaming to a cluster will work. I understand Berkeley has a ProCurve 3400 switch, but they have not yet verified it's performance. I am wondering if the ProCurve 2900 is a better (or at least good) choice, and would appreciate any comments/experiences people have.

I guess that's about it for now.

Best, Glenn

Tim Ikin, Jodrell Bank

Dear All,

As you might recall I was keen to get our old Serendip card doing useful work at Jodrell Bank - and so far development for this has been going very well. I now have sampling going stably with well behaved control and 1PPS markers stamping onto the data correctly. This has been done in Verilog which I am finding a very nice language to work with. In the next few weeks I hope to connect this up to the Lovell and then it will be over to other people to develop code on the dual-quad processor server to accept the samples and perform incoherent dedispersion on it.

(Incidentally, Dan, we would be interested to purchase another quadrature downconverter module - we have one, but need another for the second polarisation.)

Since the workshop my boss, Michael Kramer, is very interested in Casper hardware in view of supporting multibeam observations on Parkes, and Efflesburg telescopes. As you may know we have 13 beams on Parkes at 300MHz, and there will be 7 beams on Efflesburg at 100MHz. Dan has quoted us for supply, and various levels of firmware development. This will be based on iBOB's with dedispersion being on Linux servers. Personally with my other hat I am heavily involved in the UK SKA effort, but I intend to contribute to this effort as much as possible.

Kind regards, Tim Ikin

John Ford, NRAO

Hi All.

Our efforts have been mostly in planning and procuring hardware and software. We have an additional development system coming on line. We also have a Fujitsu XG-700 12 port 10 GbE switch, and HP Procurve 10 Gb-> 1 Gb switch, a dual core, dual processor Opteron server, and a 4 processor dual core Opteron server, all with Myricom 10 GbE cards. Joe Brandt has purchased several computers to use in a GPU cluster (I know. Heresy!) We also have a BEE2 and 3 iBOB's with iADC's for development work. All this hardware should be operational within a week or 2 from now.

Glen Langston, Brandon Rumberg, and Patrick Brandt, along with help from the CASPER group, have designed, built, and connected to our 43m telescope a transient event detector using iADC's and iBOB's.

Our new Pulsar Machine project, which we have named GUPPi, is underway with design work for the software and hardware proceeding. See

https://wikio.nrao.edu/bin/view/CICADA/NGNPP

We also have another BEE2 on order that belongs to West Virginia University, but problems with the board at SAE Materials have prevented WVU from getting their hands on it.

We optimistically expect to put out the first version of our machine ( a pulsar spigot machine replacement) in January for expert users, (i.e. people on the development team and their collaborators), and then we expect to release a machine with wideband coherent dedispersion by summertime. After that, we will make the instruments available to all observers as soon as we can, probably on the timescale of a year following successful first light.

John

Paul Demorest, NRAO

Hi everyone,

I've been investigating the use of GPUs for coherent dedispersion. This may be of interest to anyone who is considering streaming CASPER data out to a computer cluster for further processing. I have a webpage describing the results so far:

http://www.cv.nrao.edu/~pdemores/gpu/

To summarize, a current nvidia GPU can coherently dedisperse about 40-60mhz BW in real time (depends on DM, RF, channel BW as usual). I was also impressed by how easy it was to get this all running, using nvidia's CUDA programming language and libraries.

We're considering using these as a short-term upgrade to our ASP systems (Serendip5/PFB -> linux cluster), for use while the BEE2-based system John mentioned is still in development. This would reduce the number of cluster nodes (from ~20 currently to 4), and let us use the full 128mhz BW that comes out of S5 (currently we get 40-96mhz, depending on DM/etc).

-Paul

Matthew Bailes, Swinburne

Dear Paul et al.,

Currently we find a single Dell 1950 can coherently dedisperse a 64 MHz BW dual-pol IF at L-band in real time for most "realistic" DM/nu3 combinations @ 20cm. These are dual-processor quad core machines. I believe that they are good boxes for coherent dedispersion. The memory bandwidth is about 20 GB/s. I haven't tested how many pulsars we can simultanously fold, but for Ter 5 this would be a big issue. Can't see how to avoid the memory bandwidth problem when attempting real time coherent dedispersion on 30+ pulsars. Tell Scott to stop finding them.

Re GPUs, there are moves from nVidea to remove the DVI outputs and create a new product "Tesla" I think it is called. We had a rep out here 2 weeks ago showing us the public roadmap (no NDA) that will put four of the new GPUs with no DVI outputs into a single box, and the thing will put out 4 teraflops! Cuda enables you to program it in real languages, like C. I think one Tesla could coherently dedisperse 1 GHz BW in real time. They will cost about 20K. How you get the data into it is probably more of an issue than the CPU power.

Cheers - Matthew

Joeri van Leeuwen, CASPER

At Berkeley we are working on our incoherent filterbank on a single iBOB. That's the proof-of-concept #1 in our roadmap, relatively few channels + stokes detector + accumulator. Peter has just finished the initial testing in hardware. Once this design passes some more advanced tests we will e-mail it around.

Next steps are to include the 10Gbps blocks+hardware, on a path towards implementation in the ASP family of coherent dedispersion machines (roadmap #2-4); Parallel to that we plan to soon test the filterbank capabilites on the Allen Telescope Array.

For the roadmap and our status reports, please see Pulsar Machine Development At Berkeley

Kind regards, Joeri - for Peter, Don and Dan.