Transcript GridPP2 Coolaboration Meeting, June 2004

GridPP Collaboration Meeting
Networking: Current Status
Robin Tasker
CCLRC, Daresbury Laboratory
3 June 2004
Setting the Scene
Retrospective of recently complete / completing projects with relevant output
Output from these projects feeds directly into the GridPP Network Objectives
o To bring very high rate/long distance data transport to practical use in
experimental production environments (both current and LHC experiments).
o To exploit the UKLIGHT infrastructure to utilise switched dedicated circuits
between major centres including UK Tier1 and 2, CERN and FNAL
o To participate in EGEE network monitoring service developments and
deployment within the UK, and development of diagnostic engines for Grid
operations
Presenter Name
Facility Name
WP2, High Performance Networks
http://icfamon.dl.ac.uk/DataTAG-WP2/


Task 2.1, High Performance Transport
Transport applications for high bandwidth-delay connections








Task 2.2, End-to-End Inter-Domain QoS
To study the problem of Quality of Service provisioning in inter-domain
heterogeneous environments








D2.1, D2.2 delivered on time
Linux Networking Kernel Technical Report
SC2003; PFLD Workshop CERN (2003), Argonne(2004); GNEW 2004
Presentations / Publications / “How To” guides
Wide European and International Collaboration
D2.1, D2.3 delivered on time
D2.3 Supplement – TCP stacks and IP QoS
PFLD Workshop CERN (2003), Argonne(2004); GNEW 2004
Presentations / Publications / “How To” guides
Collaboration with Dante and the NRENs
Task 2.3, Advance Reservation
The evaluation of different advance reservation approaches and their
interoperability between GRID domains


D2.4, D2.5 delivered on time
Presentations / Publications / GNEW 2004
Final EU DataTAG Review, 24 March 2004
3
High Performance Transport
Achievements

End System Performance – Motherboard and Network Interface Card


Linux Kernel Performance – Operating System, Protocol Stack, Drivers




Methodology and tools to analyse and understand the performance of the
hardware; to ensure that the hardware configuration is optimal.
Linux Networking Kernel: Technical Report describing the operation of TCP.
Develop kernel “patches” to improve performance and feed back such
information to the Linux kernel developers.
Analysis and test of driver strategies.
Transport Protocols – The Internal Operation of TCP






Understanding Standard TCP – slow start and congestion avoidance; AIMD;
packet loss; RTT effect; MTU effect; TCP fairness; the role of ECN.
Implementation and demonstration of High Speed TCP; Scalable TCP; Grid DT
and H-TCP.
Comparison of the different TCP stacks to assess their effect
Forensic examination of TCP during operation using Net100/Web100
Experience of FAST TCP from Caltech across DataTAG
I2 LSR: broken 5 times in 2003, currently at 5.64 Gbps with IPv4 between
CERN and Los Angeles (10,949 km) and at 4 Gbps with IPv6 between CERN
and Phoenix (11,539 km )
Final EU DataTAG Review, 24 March 2004
4
Achievements of the project
 Built a leading edge, multi-domain, QoS enabled network running at 2.5 Gbit/s
 Proved “Carrier” and “Campus” class router equipment
 Demonstrated use of QoS in scientific environment
 A managed bandwidth solution for SuperJANET based on MPLS
 High performance data transport achieved and understood (leveraged DataTAG)
 24 hours > Gbit/s
 800 Mbits/s disk to disk for real data
 Demonstrated edge control of QoS (with sister project “GRS”)  see demo
 Successful External activities
 DataTAG
 QoS work with Geant/DANTE
 UKLIGHT
 RealityGrid and SC2003
 Dissemination
 Technical reports
 Conferences and papers
 A dissemination workshop in July
BaBar Data Transfers from Manchester to RAL
BaBar @ Manchester
Current transfers of 900 GBytes of BaBar
data from RAL to Manchester ~40 Hours
100 Mbit/s
The two rates are BBFTP “verbose and “non-verbose” mode
MB-NG
MB-NG does the same transfer in ~10 Hours
400 Mbit/s
CCLRC ATLAS Centre
Achievements
Network Monitoring
 Definition
and deployment of a scalable Grid network
monitoring architecture (large successful usage of RGMA)
 Probe
Coordination Protocol deployed, scheduling all network
measurements between NMs
 Central
MySQL Archive hosting all network metrics and
GridFTP logging for online analysis
RTT between sites
GridFTP analysis
Total Grid Traffic
Exploitation
 Firstly,



WP7 has largely worked with Grid standardization bodies:
Large participation to GGF areas and working groups, and in
particular with Grid High-Performance Networking (GHPN-RG) and
with Network Measurement (NM-WG).
Work towards a first prototype of an OGSI (Grid service) based
measurement architecture
Close participation in the GLUE initiative
 Secondly,
products developed by WP7 have been deployed on
EDG testbeds but also successfully in other grid environments:


The WP7 Network Monitoring Infrastructure, based on standard and
homemade measurement tools: CrossGrid, LCG
MapCenter, a grid monitoring and visualization tool: DataGRID,
DataTAG, CrossGrid, LCG, GridIreland, PlanetLab, L-Bone, Atlas Grid,
E-Toile, CEOS Grid, Nanyang Campus Grid, etc…
Year 1 Achievements
Purpose: “…design and deploy an infrastructure for network performance
Monitoring within the UK e-Science community.”
Consistent view at each site will allow easy navigation across the infrastructure.
http://gridmon.dl.ac.uk/gridmon/
"Start page" at each site shows the centre's connectivity to other UK sites based on
the last set of tests. Sites are “traffic light” colour coded, and floating text displays
the packet loss (%) .
Clicking on a site takes the user to the input form for that site, from where they can
select the remote sites, metrics and date range of interest.
Clicking "View Plot" produces the corresponding data plot.
Year 2 Achievements
1. Within the GGF Network Measurements Working Group (NM-WG):
Defining XML schemas for
requesting tests and historic data,
and publishing network
measurements.
Network
Monitoring
Service
test request
(request schema)
Aims to standardise
communication, and use XML, for
web services and OGSI/WSRF
model.
tests results
(publication schema)
2. Within the GGF Grid High Performance
Network Group (GHPN-RG)
Plan to model the network as a Grid
resource, forming an integrated stack. This is
to be achieved through a set of network
sub-services, inc. a network monitoring
service
GridPP-2 Network Objectives
1. To bring the technology of very high rate/long distance data transport to
practical use in experimental production environments (both current and
LHC experiments); and to demonstration of one or more UK HEP
experiments being able to regularly transport data at rates in excess of
500 Mbit/s and preferably 1 Gbit/s.
2. To exploit the UKLIGHT infrastructure to utilise switched dedicated circuits
between major centres including UK Tier1 and 2, CERN and FNAL.
3. To participate in EGEE oriented network monitoring service developments
and deployment within the UK, and development of diagnostic engines for
Grid operations
4. To maintain the strategic relation which HEP holds with all relevant major
Network authorities globally.
5. To provide PPNCG support and other work as specified
Presenter Name
Facility Name
Practical Stuff for High Performance
1. A workshop….
2. A Cookbook, i.e. a web presence, building on the DataTAG work
with the “How to…..” guides and the Linux Networking Kernel technical report
3. Practical Examples: extending from what’s possible in real life, e.g. MB-NG
Presenter Name
Babar work, Reality Grid and SC2003, now and in the future
Facility Name
Practical Stuff for High Performance
ESLEA: Exploitation of Switched Lightpaths
for eScience Applications
The following applications can benefit
immediately from UKLight:
HEP :Particle Physics Data transfer
VLBI: Radio Astronomy Very Long Baseline Interferometry
HPC: Computational steering and Visualisation
eHealth: Oncology Remote Visualisation
We propose a coordinated approach across these disciplines, whereby resources
and expertise is shared, and the benefit gained from the leverage of other projects
and resources accrues to all disciplines.
Network Monitoring for GridPP
Already working with
To provide infrastructure of geographically distributed monitoring nodes within
the GridPP community
To define standardised interfaces for machine access to monitoring nodes using
Grid/Web services. Continuing the work within the GGF NM-WG and GHPN-RG,
o to develop the network monitoring service based on standardise communication,
and XML for the web services and OGSA/WSRF models; and
o to model the network as a Grid resource to include a network management service
To develop requirements specification through consultation with the GOC
e.g. e-Science, and EGEE, for fault detection & analysis techniques and to
participate in the Internet2 piPEs project to develop a “diagnostic engine”
Presenter Name
Facility Name
Questions?
Presenter Name
Facility Name