Transcript WP7 - WP6

WP7 (Network) status &plans
Paris
6th of March 2002
Pascale PRIMET
INRIA/ RESO - ENS Lyon
[email protected]
1
WP7 specificity
• In testbed0 and testbed1 the networking
functionality was here (<= IP technology)
• WP7 is in charge of a shared resource called
“Network Element”
–
–
–
–
Quantifiable => network monitoring
Efficient => advanced transport protocols
Flexible => differentiated network services
Secure => firewalls/ encrypted channels?
Pascale PRIMET/ INRIA
2
Logical view of the Grid Network
Ressource = CE
(computing element)
ou Ressource = SE
(storage element)
Pascale PRIMET/ INRIA
3
« Physical » view of the Network
Public Network
No security
No predictable performances
No control on the traffic
The flat INTERNET
R = CE ou R = SE
Pascale PRIMET/ INRIA
4
WP7 aims
=> WP7 is helping Network Service Provider
in designing specific E2E network services
by mapping the application/middleware
requirements to network services (QoS,
Multicast, security…)
<=WP7 is helping A/Mw in getting E2E
performance : measurement, problems (TCP
tuning, hardware limitations….)
Pascale PRIMET/ INRIA
5
What is a “ Network Element”?
Network Element
associated to the
SiteA
SiteA
SiteB
SiteD
SiteC
Pascale PRIMET/ INRIA
6
How to characterize a “NE”?
• a NE is belonging to some Site
• A NE is a “shared resource” that permit
communication over a WAN interconnexion
• a NE is a set of oriented IP links
• a NE is characterized at IP level by:
•
•
•
•
•
•
A loss rate mean/max/average…
A loss pattern
A delay mean/max/average…
A delay pattern
A nominal rate (or TCP throughput) mean/max/average…
An aggreed QoS….(SLA)
• WP7 is discussing this concept, trying to clarify and model
it for the ATF
Pascale PRIMET/ INRIA
7
WP7 specific plan for 2002
T7.1 : Collaboration with Dante/NRENs/SEQUIN
– Pilot services test (QoS, multicast)
– Dedicated machines in GEANT PoPs
T7.2 : QoS and advanced services
- Requirements studies and test (WP10)
- Reliable Multicast
- High performance transport protocol (TCP/nonTCP)
T7.3 : Network Monitoring Architecture
– Refine the model with WP2/WP3
– Prepare a plan to analyse monitoring data
T7.4 : Security => pls see Dave Kelsey talk on Friday morning
Pascale PRIMET/ INRIA
8
WP7 Strategy
• Year1:
– Network requirements studies
– Network connectivity/capacity evaluation
– Initial network monitoring architecture proto/test
• Year2:
– Optimize the Monitoring Architecture
• (visualization, measurement scheduling, data analysis…)
– Performance optimization
– Security design
– Experiments and developments with Net providers,
with other network projects (Geant, Sequin? DataTag)
Pascale PRIMET/ INRIA
9
WP7 strategy
• Year2 : study/test; Year3: provide
–
–
–
–
Network forecasting tools/API (perf opt)
Grid traffic monitoring tools (accounting)
High throughput services/API? (perf opti)
Reliable multicast service/API? (new service)
Pascale PRIMET/ INRIA
10
Network Monitoring Architecture
Network managers
P_RTPL
MapCenter
Resource Broker
P_NWS
Middleware
LDAP
Publication
Forecaster
Analysis/ processing
Data processor
Storage
Data Collector
Raw
Mesurement
PingEr
IPerf
GridFTP
SNMP
Pascale PRIMET/ INRIA
RTPL
…
11
Collaboration with GEANT
• E2E :Close participation to pilot services
– Test of IP Premium service/WP10
• In Backbone :
– Use of dedicated machines in GEANT POPs
• Amsterdam, Geneva, London
– Tests of high throughput transfers
– Test of IP multicast for Reliable Multicast
– Sharing WP7 monitoring and DANTE monitoring data
Pascale PRIMET/ INRIA
12
WP7 and other collaborations
• WP7 and DataTAG
– QoS service study and experiment
– High Throughput study and experiment
– Network monitoring and measurement
Pascale PRIMET/ INRIA
13
WP7 and other WPs interactions
• WP1 – WP2 : monitoring/forecasting for
scheduling/ optimization
• WP3 : NetMon/R-GMA test /MapCenter
integration
• WP6 : provisionning/ firewall config/ LAN
config/Netmon deployment…
• WP8-9-10 : VO links and traffic monitoring, QoS
and perf requirement studies/test
Pascale PRIMET/ INRIA
14
WP3/WP7/WP4 : Monitoring
• 14H30 – 16H : 5th of March /EDG Paris
conference
Netsaint presentation (INFN)
• Discussion : how GUI can be interfaced with MapCenter,
R-GMA and WP4 sensors?
• Coordination of the Network Monitoring activity in EDG
- WP3/WP7 interaction:
- GRIDftp logs
•
Pascale PRIMET/ INRIA
15
WP7/WP2/WP1: optimization
• Tuesday afternoon 16h-18h :
• Presentation of the Query Optimization API (WP2)
• Presentation of the replica optimisation based on an
economical model
• Algorithm for network optimization (network cost
function) (WP7)
• Discussion : coordination of this activity
 WP7 will provide an API for a NetworkCost function
 Defining a « site ID » - to map CE or SE IP address to some IP
address/hostname representing the site.
 Deadline : end of May
Pascale PRIMET/ INRIA
16
Resource sharing
• A long term debate in the Internet community: QoS-IP
• 3 strategies:
• No garanties => Best Effort
– Measure, monitor the perf and adapt
– Communication/computing/storage
• Strict Guaranties => Reservation
– throughput, CPU, disk
• Statistical Guaranties => Prioritisation
– Connexions, jobs, users….
• Which type of guaranty you really need,
• how the solutions scales, what is the cost
• What are the advantages
theINRIA
drawback of the
Pascaleand
PRIMET/
solutions…
17
Resource sharing
•
•
•
•
Which type of guaranties do you really need?
Do the solution scales ?
What is the cost ?
What are the advantages and the drawback of each
solutions?
• In the Internet, we are working since more then ten years
on the subject….
• We still have only Best Effort…and adaptive
protocol/application like TCP…
Pascale PRIMET/ INRIA
18
Pascale PRIMET/ INRIA
19
Pascale PRIMET/ INRIA
20
Monitoring Tools
• Monitoring of Round Trip Time and Packet Loss
– edg-pinger (integrated in release 1.0)
– RIPEncc one way loss and delay (specific links)
• Monitoring of throughput (release 1.1 or 1.2)
– edg-iperf
– edg-rtpl
• DataGRID status visualization
– Mapcenter (collaboration with GLOBUS and WP3)
– Other tools evaluation (Netsaint....)
Pascale PRIMET/ INRIA
21
Forecasting tools
• WP7 is exploring and improving the NWS
(Network Weather Service) in close
collaboration with NWS designer (Rich
Woski)
• Requires to work closely with :
– WP1 : integrate network forecasts in cost
function for scheduling of jobs
– WP2 : for replication of files
Pascale PRIMET/ INRIA
22
Quality of Service
QOS in DataGRID has two Different goals :
• Guaranty high throughput for all bulk
transfers. Work ongoing with DANTE
• Guaranty prioritization for specific
application needs. Work ongoing with WP8,
WP9 and WP10 to identify precisely such
applications.
Pascale PRIMET/ INRIA
23
TCP Throughput Optimization
WP7 is working on different solutions :
• TCP mechanisms optimization
– Tests of applicability of new mechanisms
• Use of QoS solutions
– diminution of Packet Loss
Pascale PRIMET/ INRIA
24
Information System
• Close Collaboration with WP3. WP7 is
testing the R-GMA
• An LDAP backend is existing for edgpinger and other tools.
Pascale PRIMET/ INRIA
25
Contacts
•
•
•
•
•
•
•
•
•
•
WP7 manager: Pascale Primet
WP7 deputy : Peter Clarke
ATF and Integration team : Franck Bonnassieux
Collaboration with Dante: Tiziana, Richard
Monitoring : Robin Tasker, Cees DeLaat
Information System : Peter Clarke
QoS : Tiziana Ferrari
High Throughput : Richard Hughes Jones
Forecasting : Robert Harakaly
Pascale PRIMET/ INRIA
Security : Dave Kelsey
26
We aggreed
• The emphasis of WP7 work is now on task T7.2.
• Propose evaluating Scavenger and assured BW
service for TCP transfers with Dante
• to request access to the Dante infrastructure until
July 2002
• Franck will receive all bugs for the WP7 tools and
send them on to the person responsible
• To re-start the monitoring for WP7 test sites.
• Richard and Tiziana offered to run the experimental
TCP code
Pascale PRIMET/ INRIA
27
Actions list
• Action Richard: put new Technical Annex on the web.
• Action Pascale, Peter: Define the CERN EU demonstration
and call in experts as required.
• Action all: Please give any presentations/papers on WP7 to
Pascale so that they may be recorded under DataGrid
dissemination.
• Action Pascale& Peter: To submit a paper on monitoring
deliverable D7.2 for Computer Communications special journal
issue on Grid computing - deadline 1 Mar 02.
• Action Pascale: Draft an invited paper on WP7 work to be
presented at Terena by end of April.
• Action Richard: Put in link to the ccwp7.in2p3.fr site on the
public WP7 site
• Action Tiziana: check Cisco version and give the Queue
monitor command
Pascale PRIMET/ INRIA
28
Action list
• Action Rich discuss possible use of UDPmon with Sylvain
• Action iGrid2002
• Action Tiziana: Ask for a suitable (eg 2-3 month) loan of
Juniper 1GE interfaces
• Action Roberto Dante to check for space and free FPC in the
M160 and provide a list of PoPs that we might be able to use.
• Action Richard Tiziana Pascale: Supply a proposal to Dante
10 days prior to the next Policy Committee,
• Action Tiziana : Supply Dante with more details of IP Premium
users as input to the Sequin workshop.
• Action All : send update information on the maps to Frank
• Action Robin, Jules: develop the monitoring workplan for the
next meeting including collaboration with WP3.
Pascale PRIMET/ INRIA
29
Action List
• Action ALL: take the current .rpm as delivered and install it at
the WP7 test sites.
• Action Robert : circulate a detailed proposal on a possible
protocol for scheduling.
• Action Tiziana : circulate the WP1 paper
• Action Pascale: try to organise a cross WP meeting on
monitoring / possible use of NetSaint in Paris.
• Action Franck : bring NetSaint to attention of ATF
• Action Tiziana : Create a list / outline of the report with subeditors for the sections
• Action Rich/Tiziana : Make a roadmap of the 7.2 work – who
does what.
Pascale PRIMET/ INRIA
30