Download Poster - The University of Manchester

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Low-voltage differential signaling wikipedia , lookup

RS-232 wikipedia , lookup

Asynchronous Transfer Mode wikipedia , lookup

Recursive InterNetwork Architecture (RINA) wikipedia , lookup

Cracking of wireless networks wikipedia , lookup

Internet protocol suite wikipedia , lookup

Wake-on-LAN wikipedia , lookup

RapidIO wikipedia , lookup

Buffer overflow wikipedia , lookup

Buffer overflow protection wikipedia , lookup

Real-Time Messaging Protocol wikipedia , lookup

Deep packet inspection wikipedia , lookup

TCP congestion control wikipedia , lookup

IEEE 1355 wikipedia , lookup

UniPro protocol stack wikipedia , lookup

Transcript
Real Time Data Transfer for
Very Long Baseline Interferometry
Simon Casey, Richard Hughes-Jones, Stephen Kershaw, Ralph Spencer, Matt Strong
The University of Manchester, UK
European e-VLBI
Gbit link
Chalmers
University
of
Technology,
Gothenburg
Metsähovi
Very Long Baseline Interferometry (VLBI) is an aperture synthesis technique that utilizes
radio telescopes from around the world, combining astronomical data in order to achieve
high angular resolution observations. The telescopes observe the same cosmic radio
source simultaneously and the use of stable atomic clocks means that signals can be
added coherently to obtain the signal out of the noise.
Onsala
Sweden
Jodrell Bank
UK
Gbit link
Torun
Poland
The interferometer technique allows signals from each pair of telescopes to be multiplied
together in a correlator to give Fourier components. These can be transformed, and nonlinear algorithms applied to produce images of the sky enabling study of the angular
structure of radio sources at resolutions better than a millarcsecond.
The sensitivity, or signal to noise ratio, is proportional to B where B is the bandwidth and
τ the integration time, and all state-of-the-art observations are noise limited. Bandwidths of
several hundred MHz are now in common use in modern telescopes. Multi-bit sampling at
the Nyquist rate can therefore give user data rates of 1 Gbit/s or higher.
Dedicated
Gbit link
Dwingeloo
DWDM link
Medicina
Italy
Thus Bandwidth is as important as integration time: we need as many Gbit/s as we can get.
Part of the VLBI network in Europe
Transporting VLBI Constant Bit Rate Data with TCP/IP
TCP is a byte-stream transport layer protocol that guarantees reliable, in-order, and non-duplicated delivery data from sender to receiver. It uses acknowledgments (ACK) from the
receiver to slide a window (Cwnd) over the data to regulate the transmission rate. If a packet has been lost, TCP interprets this as congestion on the network, and the standard (New
Reno) congestion avoidance algorithm decreases the window by half and then slowly increases it by one packet per round trip time. The packet re-transmission and the decrease in
throughput delay the delivery of data to the receiving application as shown in the centre plot below. For e-VLBI this is undesirable and can lead to loss of correlation. However, if
incoming CBR data can be stored in the TCP socket buffer during the loss Effect
event
be sent on the link faster than the CBR rate, then catch-up of the data delivery times is
of lossand
rate on data
message can
arrival time
50
possible.
Drop 1 in 5k
45
Drop 1 in 10k
40
Drop 1 in 20k
Drop 1 in 40k
No loss
Time / s
35
30
Data delayed
25
20
15
Timely arrival
of data
10
Moving VLBI data with UDP/IP
Vlbi_control
vlbi_recv
7
8
9
10
4
x 10
Effect of loss rate on message arrival time.
TCP buffer 1.8 MB (BDP) RTT 27 ms
Increase in throughput allows message catch-up.
TCP buffer 160 MB RTT 15.2 ms
At the sender, the input thread either reads data from a file or generates
random data and places these in to the ring buffer. The output thread
encapsulates the VLBI data in a UDP/IP packet together with an application
header containing sequence number which increments by 1 for each packet
sent.
Manchester - JIVE, NL (UKLight)
1000
1
800
0.8
600
0.6
400
0.4
200
0.2
0
0
2000
Input thread
8000
10000
12000
0
14000
Memory
Ring buffer
In December 2006, a 3 station e-VLBI experiment was emulated by
simultaneously transmitting data from 3 locations over the GÉANT2 network
Send thread
Receive thread
into PCs at JIVE. The achieved throughputs and packet loss are shown in
Architecture of the VLBI-UDP Program the plots to the right. The absence of packet loss clearly shows the superior
performance of the UKLight lightpath when compared with the packet
switched production network.
UDP Data
1
800
0.8
600
0.6
400
0.4
200
0.2
0
0
2000
4000
6000
8000
10000
12000
Bologna - JIVE, NL (Packet switched)
1000
1
800
0.8
600
0.6
400
0.4
200
0.2
0
0
2000
4000
6000
8000
10000
12000
Time during the transfer s
3 simultaneous flows into JIVE
Throughput and Packet Loss
RR001 The First Rapid Response Experiment
(Rushton Spencer)
The experiment was planned as follows:
 Operate 6 EVN telescopes in real time with observations on
29th Jan 2007
 Correlate and analyse the results
 Select the sources for follow up observations
 Observe the selected sources on 1 Feb 2007
The experiment worked – we successfully observed and analysed 16
sources weak microquasars) read for the follow up run, but we
found that none of the sources were suitably active at that time.
Microquasar Cygnus X-3 (10 kpc)
(a) on 20 April and (b) on 18 May 2006. The source
as in a semi-quiescent state in (a) and in a flaring
state in (b), The core of the source is probably
~20 mas to the North of knot A. (Tudose et al.)
(a)
0
14000
Time during the transfer s
e-VLBI Science
Microquasar GRS1915+105 (11 kpc) on 21
April 2006 at 5 Ghz using 6 EVN telescopes,
during a weak flare (11 mJy), just resolved in
the jet direction (PA140 deg). (Rushton et al.)
g
1000
% Packet loss
At the receiver, the receive thread places incoming packets directly into the
next position in the ring buffer. The sequence number is read from the
header and this reveals whether the packet is at the correct position in the
buffer. If the sequence number increment is not equal to 1, then the packet
is moved forwards or backwards in the buffer as appropriate.
Output thread
Ring buffer
6000
Manchester - JIVE, NL (Packet switched)
Wire Rate Mbit/s
Disk
4000
Time during the transfer s
Wire Rate Mbit/s
Memory
g
4
5
6
Message number
Control thread
TCP Control
Disk
3
(b)
The ESLEA UK e-Science project is funded by the EPSRC, PPARC and MRC Research Councils
This work was performed in collaboration with the EXPReS project, EC FP6 contract number 026642
0
14000
g
Control thread
2
% Packet loss
vlbi_send
1
% Packet loss
With packet loss, TCP decreases the rate.
TCP buffer 0.9 MB (BDP) RTT 15.2 ms
0
Wire Rate Mbit/s
5