EXPReS SA1: at the half-way mark Arpad Szomoru JIVE
Outline Progress in 2007 Going intercontinental The next 18 months Slide #2
e-vlbi science/test runs 17 e-vlbi science projects accepted since 2006: 2 failed (in early 2006) 8 active binary systems (Algol-type, X-ray or gamma-ray binaries); 4 of these were ToO projects, 3 part of multi-wavelength campaign 1 adaptive observation of 16 X-ray binaries (no detections..) 4 determination of compactness of calibrator or target 1 spectral line run 1 supernova ToO Rapid access to EVN provides clear benefit to users (important for calibrator/multi-wavelength projects) Follow-up observations of bursting transients were only a moderate success in 2006; two weeks delay between proposal submission and observations is too long e-vbli observing policy was changed in early 2007, to allow quick response to triggered projects (in 24 hours). Slide #3
e-vlbi: operational improvements Overall improvement in first 18 months: Robustness Reliability Speed Ease of operation Station feedback Minimizing data loss by careful scheduling Increase of production data rate from 128 Mbps to 256 Mbps 6-station fringes at 512 mbps Inclusion of Metsähovi and Medicina telescopes First successes with Shanghai and Australian telescopes, Arecibo and TIGO Slide #4
e-vlbi results Slide #5
Sharply reduced time to publication Data processing took 1-2 weeks with first images within 48 hours Publication took less than 2 months Slide #6
e-vlbi: data rate improvements Number of telescopes @ data rate 1024 896 Data rate (Mbps) 768 1 640 512 1 3 2 6 5 384 Aggregate data rate 256 6 4 3500 128 3000 0 Aug-04 Feb-05 Sep-05 Mar-06 Oct-06 Apr-07 Nov-07 Jun-08 2500 Data rate (Mbps) Jun-03 Jan-04-128 2000 1500 1000 500 0 Apr-04 Sep-04 May-05 Sep-05 Oct-05 Nov-05 May-06 Aug-06 Dec-06 Feb-07 Jun-07 Slide #7
The EVN MarkIV correlator Custom-made hardware, ~500000 lines of C++ code Designed and built on tape technology, only recently adapted for disk-based recording Slide #8
Work packages Standard e-vlbi correlator mode: Wall-clock time vs ROT Centralized control Rapid response/too functionality: Real-time monitoring/feedback/analysis On-the-fly adjustment of observational/correlator parameters Improved amplitude calibration Data transport issues Lightpaths: guaranteed (high) bandwidth Transport protocols (modified TCP, UDP-based, VSI-e ) Hardware issues New control computers (reliability, speed) Replacement of SUs (cause of much misery) Networking hardware (router, interfaces) Slide #9
Hardware New control computers Solaris AMD servers, redundant power supplies Interchangeable for maximum reliability Cut down dramatically on (re-)start time Removal of single points of failure Mark5 upgrade Original Mark5s woefully under-powered (Pentium PIII ) New dual processor motherboards, memory, power supplies Immediate result: Mc and Tr sustain 512 Mbps Mark5A B: new streamstor cards, new serial links (optical and coax), Correlator Interface Boards Elimination of Station Units, VSI compliance Slide #10
Upgrade of network at JIVE Slide #11
Tools for feedback and streamlining Public Status Rapid feedback Streamlined processing Adaptive observing Slide #12
The road to China. Slide #13
The APAN demo challenges Australian telescopes EXPReS demo only planned for October Need for physical bandwidth Data format conversion, playback at correlator LBADR interface Suitable transport protocol? Chinese telescopes Only Shanghai available Still no Mark5A with Linux 2.6 kernel, Reno TCP not an option, no working UDP-Mark5A European telescopes Torun willing, but under maintenance Medicina willing, but only formatter data UK possibly one dish Westerbork one dish Slide #14
e-vlbi to China Connection Sheshan (near Shanghai) JIVE (via HongKong, Chicago). RTT 360 ms, real distance 8800 km, path length 27500 km Slide #15
e-vlbi to China: Connectivity tests Sheshan-JIVE In practice: a sizzling 22 Mbps Slide #16
Hacking time Investigation by Harro Verkouter of UDP on Mark5A After discussions with Richard Hughes-Jones and Simon Casey, realization that serious hacking was needed Extra-simple UDP variation, inspired by vlbi-udp: no replacement of missing data, out-of-order packets are chucked out Has saved the Chinese part of the APAN demo Simon is catching up now Slide #17
And other solutions Home-grown version of Circuit TCP (Mudambi, Zheng and Veeraraghavan), made by Mark Kettenis using pluggable TCP congestion avoidance algorithms After additional debugging by Chris Phillips, 512 Mbps over lightpath from Australia, 256 over production network On-the-fly data conversion, playback on Mark5A+ (by Chris and Mark) LBADR correlator interface Three different protocols used in demo Slide #18
The results.. Slide #19
China: e-vlbi and (network )politics ORIENT: 2.5 G TEIN2: 622 M 155 M 100 M 32 M 34 M 2.5 G 50 M 1G 1G 2.5 G 622 M HKOEP SURFnet Slide #20
1 Gbps 2.5 Gbps 10 Gbps 155 Mbps
SA1: the next 18 months Main focus: Adaptive scheduling Increasing the operational data rate Improving flexibility and robustness of correlation New telescopes Enabling mixed-mode operations Inclusion of emerlin telescopes Slide #22
Specific tasks Crucial: fast/adaptive (re-)scheduling Transform EVN into a truly flexible instrument Reaction time of hours Remote control of widely different telescope control systems Safeguard local operational constraints hard/software upgrade Phase out aging Station Units, main cause of operational instability VSI compliance Slide #23
Fine-tuning of deliverables Use of WSRT synthesis data for e-vlbi calibration (absolute flux calibration, polarization calibration, source selection) Space craft tracking correlator mode On-the-fly fringe fitting Real-time download and extraction of station information Automated correlator diagnostics Removing/adding stations from the correlation process on the fly Investigating a 1024M sub-array Slide #24
Connectivity improvements: Work toward stable UDP solution on Mark5 Increase granularity of data streams to optimize utilization of available bandwidth Increase of sensitivity and resolution through addition of telescopes: Effelsberg, Yebes, Shanghai Improving global connectivity: South-America, Puerto Rico, China, Australia Slide #25
Global connectivity: Arecibo: recent UDP tests yield sustained 64 Mbps. New submarine cable facilities (bypassing Abilene) in near future Australia: 4 LBA telescopes connected at 1 Gbps, one 1Gbps lightpath from Sydney to JIVE. AARNET will provide a total of three 1-Gbps lightpaths for demo in October (3 * 512 Mbps to JIVE) TIGO: 64 Mbps UDP sustained with data generator South Africa: local situation improving, connection to Europe remains problematic Slide #26
e-vlbi Slide #27