DataTAG - WP3 Objectives, activities and progress by UvA: Wim Sjouw, Hans Blom, Jason Lee, Cees de Laat INRIA: Mathieu Goutelle CERN: tbd
Contents Objectives Tasks Activities Experiences with SURFnet Lambda Pointers
WP3 objectives Bulk data transfer and application performance monitoring: innovative monitoring tools are required to measure and understand the performance of high speed intercontinental networks and their potential on real Grid application.
Tasks in WP3 Task 3.1. Performance validation (month 1-12) –Create, collect, test network-tools to cope with the extreme Lambda environment (high RTT, BW) –measure basic properties and establish a baseline performance benchmark Task 3.2 End user performance validation/monitoring/optimization (month 6-24) –Use “out of band” tools to measure and monitor what performance a user in principle should be able to reach Task 3.3 Application performance validation, monitoring and optimization (month 6-24) –Use diagnostic libraries and tools to monitor and optimize real applications to compare their performance with task 3.2 outcome.
Task 3.1 Performance validation Performance Test Tools – Lambda's are a Layer 1-2 environment –use end to end traffic generating tools –not ICMP based tools since those need routers in the path –tools must work in Gigabit environment Netperf –TCP bulk transfer, UDP unusable (no shaping) –modifications made to enhance IPv6 support Iperf –TCP and UDP shaped bulk transfer –lightweight multiple streams test support –IPv6 support –modifications made to introduce TCP shaping, debugging
Task 3.1 Performance validation (cont’d) UDPmon (by Richard Hughes-Jones) –each packet contains seq. # / time stamp info receiver can deduce: # packets lost # packets out-of-order UDP bandwidth –very precise timing –not intrusive –tool is in continuous development
Task 3.1 Performance validation (cont’d) Tools under investigation which are either not released yet or incomplete: Pathload : estimate available bandwidth Magnet : network software performance Netprobe : one way delay + variance, jitter
Task 3.1 experiences using Netherlight SURFnet Lambda AMS-CHI full duplex GE 2.5 Gbps SDH, 100 ms RTT –single stream TCP: max. throughput Mbps, dependent on stream duration –similar for BSD and Linux and different adaptors –UDP measurements show effects of hardware buffer-size in ONS equipment when assigning lower SDH bandwidths, see:
Task 3.2 End user performance validation/monitoring/optimization Toolset for monitoring –Re-use as much as possible WP7 stuff –rTPL(end user performance monitoring): –MRTG (to monitor router interfaces): –SmokePing (history and statistics of rtt’s): Not so ideal tools (for the time being): –Cricket (to monitor large number of router interfaces):
Task 3.2 and 3.3 set up tool RTPL for end user performance check, can be expanded when DataTAG lambda becomes available. See: task 3.3 will most probably use netlogger, did not yet start. all info can also be found via:
Summary of status Tools have been investigated, selected and where necessary adapted to benchmark and characterize Lambda networks – New tools appear and need to be studied to add to the set Influence layer 1 and 2 infrastructure properties on TCP throughput is under study on SURFnet Lambda testbed and NetherLight Monitoring setup is underway, inclusion of WP7 toolset is next step Application performance monitoring and optimization should start soon Preparation for iGrid2002 which also gives us: –a 2.5 gig connection Amsterdam-CERN –for duration of iGrid Gbit/s connection Starlight-Amsterdam by Level3! – in Amsterdam septemberWWW.IGRID2002.ORG