Transcript WP7 Networking

WP7 Networking
Richard Hughes-Jones
GridPP meeting Feb 03 R. Hughes-Jones Manchester
WP7: Networking for Grids
 Grid
Network monitoring

Provide information for Middleware & Applications – Network Cost Function

Understand the networks we use

Provide Information for capacity planning
 Creation
of schemas and publishing the monitoring data
 Investigation

of Protocols TCP and non-TCP
Testing the work of CS groups / IETF NOT inventing
 Close
 The
technical collaboration with NRNs, DANTE and the DataTAG project
High Bandwidth High Throughput Challenge

Investigation of end Host Networking and Disk sub-systems

To show what can be achieved on production networks with:
 Multiple streams of TCP packets
 Tuned TCP parameters
 Different TCP stacks
Applying
the knowledge to the real Grid user community
GridPP meeting Feb 03 R. Hughes-Jones Manchester
Network Monitoring – Technology  e-Sci
GridPP meeting Feb 03 R. Hughes-Jones Manchester
NetworkCost Architecture
Processing
NetworkCost
Collect
And
Storage
Raw
R-GMA
Globus MDS
Archive
Distributed Data Collector
Measure
PingEr
IPerf
UDPmon
PCP
GridFTP
GridPP meeting Feb 03 R. Hughes-Jones Manchester
NetworkCost functionality
CERN
CERN
RAL
NIKHEF
IN2P3
CNAF
46,75
77,78
44,87
35,44
2,44
7,12
4,35
11,86
2,66
RAL
7,46
NIKHEF
11,13
3,25
IN2P3
5,03
10,38
6,24
CNAF
4,5
6,53
4,04
7,08
13,08
cost[][] =
getNetworkCost (SE[], SE[])
FileSize= 11 MB
GridPP meeting Feb 03 R. Hughes-Jones Manchester
High throughput transfer challenges
 Large
amounts of data have to be transferred between Mass Storage Systems
and CEs in Europe (and world wide!)
 EU
demonstration sent HEP data from CERN to NIKHEF/SARA at high rates
 It
was to show what can be achieved with:
 Multiple streams of TCP packets
 Tuned TCP parameters:
Interface txqueuelen 2000
TCP buffer size to match the BW * rtt
 Different TCP stacks:
Standard TCP
Fast TCP
Scalable TCP
 Fair sharing between stacks
This
highlights the results of close technical collaboration with NRNs,
DANTE and other projects: DataTAG, Mb-NG,
UK- Star- Nether- Light
GridPP meeting Feb 03 R. Hughes-Jones Manchester
Demo Setup for the EDG Review
 Shows
data transfers from Mass Storage system at CERN to Mass
Storage system at NIKHEF/SARA
 Disk
 All
sub-system I/O bandwidth of ~70 MB/s
systems have Gigabit Ethernet connectivity
 Use
GridFTP and Measure disk to disk performance
SurfNet
GEAN
T
NIKHEF
CERN
GridPP meeting Feb 03 R. Hughes-Jones Manchester
Demo Consisted of:
Data over TCP Streams
Raid0
Disk
Node Monitoring
GridFTP
GridFTP
Site Monitoring
GridPP meeting Feb 03 R. Hughes-Jones Manchester
Raid0
Disk
Dante Monitoring
European Topology: NRNs, Geant, Sites
Sara & NIKHEF
SURFnet
SuperJANET4
CERN
GridPP meeting Feb 03 R. Hughes-Jones Manchester
Throughput on the day !
The view from GÉANT – with thanks to Dante
GridPP meeting Feb 03 R. Hughes-Jones Manchester
Some Measurements of Throughput CERN -SARA
TCP
 Average
 Users
200
0
1043509370
see 5 - 50 Mbit/s!
 Average
300
100
Throughput 167 Mbit/s
High-Speed
400
TCP
Throughput 345 Mbit/s
1043509470
1043509570
Time
1043509670
Hispeed TCP txlen 2000 26 Jan03
500
400
300
200
100
0
1043577520
1043577620
1043577720
Time
1043577820
Scalable TCP txlen 2000 27 Jan03
 Average
Throughput 340 Mbit/s
400
300
200
100
0
1043678800
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
Out Mbit/s
1043679200
In Mbit/s
Recv. Rate Mbits/s
TCP
II/f Rate Mbits/s
500
Scalable
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
Out Mbit/s
1043577920
In Mbit/s
Recv. Rate Mbits/s
Standard
Out Mbit/s
2
In Mbit/s
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
1043509770
Recv. Rate Mbits/s
GByte file transfers
I/f Rate Mbits/s
1
the GÉANT Backup Link
I/f Rate Mbits/s
Using
Standard TCP txlen 100 25 Jan03
500
1043678900
1043679000
Time
GridPP meeting Feb 03 R. Hughes-Jones Manchester
1043679100
What the Users Really find:
– RAL using production GÉANT
 CMS
 50
Tests 8 streams
Mbit/s @ 15 MB buffer
 Firewall
100 Mbit/s
hroughput Mbit/s
CERN
CERN -RAL 12 Dec 02
90
80
70
60
50
40
30
20
10
0
0
NNW
1
10
20
30
time 0.5 hr
– SJ4 Access
Gbit link
GridPP meeting Feb 03 R. Hughes-Jones Manchester
Total Rate
Rate/Stream
40
50
WP7 High Throughput Achievements
Close
Collaboration with Dante
“Low”
layer QOS testing over GEANT
 LBE
 IP
iGrid
premium
2002 and ER 2002 : UDP with LBE
 Network
performances evaluation
EU
Review 2003 : application level
transfer with real data between EDG sites
 proof
of concept
GridPP meeting Feb 03 R. Hughes-Jones Manchester
Conclusions
More
 ie
research on TCP stacks and its implementation is needed
HEP-style applied research -
Continue
the collaboration with NRNs & Dante to:
 Understand
 Learn
the behavior of National networks & GEANT backbone
the benefits of QoS deployment
WP7
is taking the “Computer Science” research and knowledge of the
TCP protocol & implementation and applying it to the network for real
Grid users
Enabling
 EDG
CE
Knowledge Transfer to sysadmins and end users
release 1.4.x has configuration scripts for TCP parameters for SE and
 Network
 Work
tutorials for end users
with users – focus on 1 or 2 sites to try to get improvements
GridPP meeting Feb 03 R. Hughes-Jones Manchester