GridPP Collaboration Meeting Nov 2001

1
Network Monitoring and GridPPRichard
Hughes-Jones, University of Manchester6 November
2001

• DataGrid WP7
• MB NG
• DataTAG
• GridPP

2
DataGrid WP7 Networking

• Active and proceeding well, Meetings
• Oxford DataGrid meeting Jul 01
• CERN 18 Sep 01
• Frascati DataGrid meeting Oct 01
• Provisioning, Reports on
• DataGrid network requirements and current
  infrastructure
• Use of IP ports for TestBed1
• Monitoring (Robin Tasker)
• High Performance High Throughput (Richard
  Hughes-Jones)
• QoS and Bandwidth Reservation (Tiziana Ferrari)
• Secuity (Dave Kelsey)

3
Network Monitoring

• Several tools in test plugged into a coherent
  structure
• PingER, RIPE one way times, iperf, UDPmonRE,
  rTPL, GridFTP, and NWS prediction engine
• continuous tests for last month to selected
  sites
• DL Man RL UCL CERN Lyon Bologna SARA NBI SLAC
• Discussions this week at WP7 in Amsterdam
• The aims of monitoring for the Grid
• to inform Grid applications, via the middleware,
  of the current status of the network input for
  resource broker and scheduling
• to identify fault conditions in the operation of
  the Grid
• to understand the instantaneous, day-to-day, and
  month-by-month behaviour of the network provide
  advice on configuration etc.
• Report written on LDAP scheme for publishing the
  network information

4
Network Monitoring Architecture
LDAP Schema
Grid Apps GridFTP


Backend LDAP script to fetch metrics Monitor
process to push metrics
PingER (RIPE TTB) iperf rTPL NWS etc
Local Network Monitoring Store Analysis of Data
(Access)
local LDAP Server
Grid Application access via LDAP Schema to -
monitoring metrics - location of monitoring
data.
Access to current and historic data and metrics
via the Web, i.e. WP7 NM Pages, access to metric
forecasts
Robin Tasker
5
How do the Grid apps access themetrics of
network monitoring?
What is the RTT viewed from UCL to RAL and QMW?
Query an LDAP server that makes use of an LDAP
Schema containing the ObjectClass RTT to find
out. How?
6
and find the details from the LDAP Schemafor
publication of the RTT metric
servicenetmon, dcucl, ouUK, ouDataGrid,
oGrid
rouqmw objectclassnetworkmonitorHost objectclass
networkmonitorRTT objectclassnetworkmonitorThrou
ghput objectclassnetworkmonitorLoss
roural objectclassnetworkmonitorHost objectclass
networkmonitorRTT objectclassnetworkmonitorThrou
ghput objectclassnetworkmonitorLoss
PingER RTT metric rTPL RTT metric
7
to allow a request to return a valid metric
8
Data RIPE 1-way TCP throughput

• RIPE 1-way time ms
• Sara ? RAL
• 20 Oct 01
• RIPE 1-way time ms
• RAL ? Sara
• TCP Iperf prediction
• Mbit/s
• UCL ? Sara

9
Data Ping UDP throughput

• PingER rtt (ms)
• dl cern
• 1000 byte packet
• Forecast
• From 20 Oct 01
• UDPmon throughput Mbit/s
• man cern
• 300 1400 byte frames

10
High Performance High Throughput

• Document produced outlining the tests to be made
• Understanding the end system HW,
• best way to monitor traffic and protocol packets,
  
• Type and effect of background traffic,
• Throughput vs rtt, throughput vs window-size,
• Throughput using multiple TCP streams, and effect
  on the Network
• GridFTP at Gigabit,
• Effect of different TCP stacks,
• Use of non-TCP protocols, and effect on the
  Network
• Show and tell demos
• Discussions this week at WP7 in Amsterdam
• Links to MB-NG, DataTAG, GGF/IETF

11
QoS and BW Reservation

• Proposed workplan presented at Oxford WP7 meeting
• Identification of traffic classes and middleware
  components requiring QoS
• Discussion with other WorkPackages definition
  of application requirements
• GridFTP ? packet loss
• Interactive applications ? packet loss, delay,
  jitter
• Piloting of IP Premium (GEANT and NRNs)
• Study of traffic differentiation and traffic
  engineering techniques
• Layer 2 MPLS VPNs (CISCO and Juniper)
• LAN WAN QoS
• Traffic clasification, marking, policing,
  ccongestion control, queue scheduling, traffic
  aggregation
• demonstration of QoS Bandwith Reservation
• Document in production outlining the tests to be
  made

12
E-science core project

• Project to investigate and pilot
• end-to-end traffic engineering and management
  over multiple administrative domains MPLS in
  core diffserv at the edges.
• Managed bandwidth and Quality-of-Service
  provision. (Robin T)
• High performance high bandwidth data transfers.
  (Richard HJ)
• Demonstrate end-to-end network services to CERN
  using Dante EU-DataGrid and the US DataTAG.
• Status approved with requested funds start on 1
  Dec 2001
• PartnersCISCO, CLRC, Manchester, UCL, UKERNA
  plus Lancaster and Southampton (IPv6)
• A technical meeting held draft spec. of the HW
  required
• Would like to use real Grid traffic maybe
• CDF UCL-RAL
• BaBar Man-RAL

13
MB NG SuperJANET Testbed
Leeds
SJ4 Dev C-PoP Warrington
SuperJANET4 Production Network
Gigabit Ethernet 2.5 Gbit POS 10 Gbit POS
SJ4 Dev C-PoP Reading
SJ4 Dev C-PoP London
RAL / UKERNA
RAL / UKERNA
14
The EU DataTAG project
DataTAG

• EU Transatlantic Gigabit project.
• Status approved for 4 M ECU start on 1 Dec 2001
• Partners CERN/PPARC/INFN/UvA. IN2P3
  sub-contractor
• The main foci are
• Grid Network Research including
• Provisioning (CERN)
• Investigations of high performance data transport
  (PPARC)
• End-to-end inter-domain QoS BW / network
  resource reservation
• Bulk data transfer and monitoring (UvA)
• Interoperability between Grids in Europe and the
  US
• PPDG, GriPhyN, DTF, iVDGL (USA)

15
DataTAG project
NL SURFnet
MREN
STAR-LIGHT
STAR-TAP
CERN
ESNET
NewYork
IT GARR-B
Abilene
Olivier Martin
16
Single stream vs Multiple streams effect of a
single packet loss (e.g. link error, buffer
overflow)
DataTAG
Streams/Throughput 10 5 1 7.5 4.375 2 9.375
2T Avg. 7.5 Gbps
10
Throughput Gbps
7 5
2T Avg. 4.375 Gbps
5
Avg. 3.75 Gbps
2.5
T 2.37 hours! (RTT200msec, MSS1500B)
T
T
T
Time
17
GGF

• Grid High-Performance Networking RG
• Talk on Lambda Networking Research, Cees de Laat,
  UvA
• BW reservation Gigabit Tests ATLAS at Michigan,
  E. Myers
• Charter fully discussed
• Network monitoring WG
• Much in common with WP7 two way exchange of
  techniques.
• GridFTP
• UK participation in protocols and a-testing with
  the Globus developers.
• LDAP Schema for network status
• Discussions with GLobus

18
Focus on The UK and GridPP

• GridPP DataGRID DataTAG MB-NG GGF will
  collaborate closely.
• GridPP has UK specific issues and includes
  experiments at
• BaBar with links to SLAC
• CDF and D0 working at Fermi
• UKQCD UKDMC (dark matter) MINOS
• PPNCG is a natural forum fro UK GridPP network
  matters
• Recent throughput problems, perceived as
  transatlantic, traced to on-campus bottlenecks.
• The PPNCG strongly encourage good liaison between
  the HEP teams and the campus networking groups.
• Compiled a picture of the access BW to the HEP
  sites

19
Connectivity of UK Grid Sites. BW to Campus, BW
to site, limit
Glasgow 1G 100M 30M?
Edinburgh 1G 100M
Lancaster 155M 100M move to cnlman at 155Mbit
Durham 155M ??100M
Manchester 1G 100M 1 G soon
Sheffield 155M ??100M
Liverpool 155M 100M 4155M soon. To hep ?
Cambridge 1G 16M?
DL 155M 100M
UCL 155M 1G 30M?
Birmingham 622M ?? 100M
IC 155M 34M then 1G to Hep
Oxford 622M 100M
RAL 622M 100M Gig on site soon
QMW 155M ??
Swansea 155M 100M
Brunel 155M ??
Bristol 622M 100M
RHBNC 34M 155M soon ?? 100M
Portsmouth 155M 100M
Southampton 155 100M
Sussex155 100M
20
Ptolemy simulation of the Grid (1)

• Ptolemy - a discrete event simulation tool from
  Berkley
• Simulation based on flows from DataGrid-7-D7.1-01
  9-1-0-NetworkRequirements.doc
• Predics the flow between Institutes and across
  SuperJANET

Paul Mealor
21
Ptolemy simulation of the Grid (2)
CERN
RAL
Bristol
SuperJANET
Liverpool
Glasgow Edinburgh
Lancaster
Birmingham
London Man
(Imperial)
Tier 3
Paul Mealor
22
Dont Forget Involvement with

• UKQCD UKDMC (dark matter) MINOS
• AstroGRID
• AccessGRID
• Grids for High Performance Computer Centres
  Edinburgh Manchester
• Lambda Switching Projects

23
European Access

• PPNCG monitoring shows packet loss and increased
  rtt from the UK to Europe.
• Due to packet loss at the TEN155-gw.ja.net
• Start of problems 12 Oct
• 10-20 loss seen by 31 Oct
• The 155 Mbit line to Dante running at 130 Mbit
• A reconfiguration Fri 2 nov helped.
• TEN155 contact ends 1 Dec replaced by Dante
  2.5Gbit access link.