The e-EVN Arpad Szomoru, JIVE EVN CBD, Bonn, May 2012
SFXC developments • Transition from Mk. IV to SFXC: • Multi-epoch disk-based projects on Mk. IV: all done • By session 1/2012, only 1 of 14 projects run on Mk. IV (probably the last) • e-EVN continues to use Mk. IV for real-time correlation • Phasing-out plan has been drafted • Depends a bit on Uni. Board first fringes EVN CBD, Bonn, May 2012
Uni. Board correlator • Uni. Board project will end June 30 • Several production runs (extra boards for Sh. AO, MPG, JIVE, ASTRON) • APERTIF shaping up: 80 to 100 boards needed • Will be used as platform for AAVP computing • Correlator development continuing • • End-to-end functionality demonstrated Cross correlations with simulated data Ironing out last issues wrt delay correction Correlator control, status and error control, handling of output data being addressed • Uni. Board 2 funded, start July 1 st • • • Build on expertise of first project Preserve personnel Strong emphasis on power efficiency 28 nm available, 20 nm “soon” Latest technology essential for SKA potential Actual design may be delayed, in order to leapfrog one generation of FPGA EVN CBD, Bonn, May 2012 3
Scaling up SFXC Current SFXC: 1 Gbps from 9 stations in real-time. • Current SFXC: 256 cores, 2 racks, 9. 5 k. W • SFXC replacement of Mark. IV: 1024 cores, 8 racks, 38 k. W SFXC as a next generation EVN correlator: additional factor of 4 to go from 16 to 32 stations, and a factor of 4 to go from 128 to 512 MHz. • SFXC as next generation EVN correlator: 16384 cores, 128 racks, 608 k. W Restricting the full 4 GHz next generation case to 2 -bit sampling, we'll need an additional factor of 8 to go to 4 GHz. • SFXC as 4 GHz next generation EVN correlator: 131072 cores, 1024 racks, 4. 9 MW EVN CBD, Bonn, May 2012
And Uni. Board replacement of Mark. IV: 1 board, 400 W Going to 32 stations and from 64 MHz to 512 MHz: Uni. Board as next generation EVN correlator: 8 boards, 3. 2 k. W And finally going from 512 MHz to 4 GHz: Uni. Board as 4 GHz next generation correlator: 64 boards, 25. 6 k. W • Clearly very suitable as heavy-duty work-horse platform • Challenge will be to implement extended functionality • Pulsar gating, multiple phase centres • SFXC and Uni. Board: complement each other very well • FPGA correlator will never be as flexible as software • SFXC will never be as powerful as FPGA correlator EVN CBD, Bonn, May 2012
NEXPRe. S • Finally, some money has been distributed • We have incredible stories to tell about bureaucracy • Advancing in year 2 • Intensifying management role in Work Package 8 • Other work-packages doing OK • Preparing 4 Gbps real-time demo (On, Ef, Yb) • Installing high-speed data servers, tests ongoing • 4 Gbps Bandwidth-on-Demand successfully demonstrated • e-VLBI important driver in development of international standard • Software, software • NEXPRe. S board will meet in Copenhagen, June 20 • Preparing for new mid-term review EVN CBD, Bonn, May 2012 6
SKA process • Prep. SKA formally ended • Uneasy feeling about role of board, now Company has taken over • JIVE currently involved in two consortia preparing Eo. I • For stage one of SKA pre-construction engineering phase • Provide expertise and feedback on various topics • SKA Signal and Data Transport and Synchronization and Timing • Network engineering, expertise with timing and frequency • Low Frequency Aperture Array • AAVP will use Uni. Board: VHDL programming, correlator design • Make sure experiences with AAVP feed back into design of Uni. Board 2 • Discussing participation in Central Signal Processor consortium • Uni. Board 2 as a viable option for SKA Phase 1 correlator EVN CBD, Bonn, May 2012 7
Some issues • Divergence in digital backends • DBBC, RDBE, KVAZAR backend, CDAS, DSN backend…. • Need to worry about compatibility • Divergence in high-speed data recorders • Mark 5 C, Mark 6, NEXPRe. S Flex. Buff (Aribox), X-Cube…. • Remarkably similar hardware • Introduction of Mark 6 in the EVN? • Need to replace ~80% of EVN disk packs • Or ~65% of disk capacity • And basically gut all Mark 5 units, install new backplane, motherboard…. • NEXPRe. S “philosophy” • All VLBI (at least partly) e-VLBI • Only ship data when necessary EVN CBD, Bonn, May 2012
Increasing operational complexity huib 08/06/10 EVN CBD, Bonn, May 2012 /28
Some more issues • Many pieces of NEXPRe. S-related software in place • • • Simultaneous recording/streaming at stations Simultaneous recording/correlating at correlator (nearly available) Tracking of disk packs, recordings on data buffers at various places Automated system monitoring, generation of warnings (email, sms) Enormously augmented functionality of Jive 5 AB: (partial) cornerturning, splitting of high-speed data streams, not just limited to Mark 5 platform anymore • Not all of this is used yet though… • Use during NMEs, ftp fringe tests would be good start • Need one control code for both e- and recorded VLBI • Extension of Jive 5 AB code to full Haystack command set/responses • Has been requested by TOG in past • Return of e-MERLIN stations (fantastic news for users) • 7 stations at a full 1 Gbps? 7 Local recorders? • Connectivity to Jb recently upgraded to 3 Gbps • Would benefit from scheme with fast buffering, partial real-time correlation, trickled e-transfer • And how about new e-members like KVN? EVN CBD, Bonn, May 2012
Скачать презентацию The e-EVN Arpad Szomoru JIVE EVN CBD Bonn
8189f75005c45d2fc514bdf16d847f68.ppt