Network connectivity between SLAC and IN2P3, Lyon, France

Les Cottrell. Page created: December 7, 2000.

Central Computer Access | Computer Networking | Network Group | ICFA-NTF Monitoring

Introduction

The IN2P3 computer center, in Lyon France is a BaBar tier A remote computing site. This means it will have a copy of all the Babar data, and will be a major site where Babar computing will be performed. As such it needs excellent connectivity/performance to SLAC and other BaBar collaborator sites. The question was therefore posed by Walter Toki of Colorado State/Babar as to how it is connected, what the Internet performance to BaBar looks like, and what are the plans.

Routes

The traceroute from SLAC to IN2P3 shows that there is about 60msec. round trip time (RTT) between SLAC and Chicago and then a further 120msec. from Chicago to Lyon, France. It is also seen that the route goes through CERN. The link from CERN to IN2P3 is 34Mbps. The CERN link to the US is an STM-1 (155Mbps) link to Chicago. The traceroute frpm IN2P3 to SLAC when compared with the traceroute from SLAC to IN2P3 shows the routes are fairly symmetric. The path characteristics measured from SLAC to IN2P3 indicate that the bottleneck bandwidth to the IN2P3 site is about 30Mbps. This bottleneck limit is roughly confirmed by bulk throughput measurements between SLAC and IN2P3.

Historical performance

Using the IEPM/PingER we can investigate the long term performance, in particular the packet loss, between IN2P3 and various PinGER monitoring sites around the world. The table below shows the loss seen between several PingER monitoring sites and IN2P3. Unfortunately there is no Pin gER monitor at IN2p# so we cannot see how sites look seen from IN2P3. The numbers in the cells are the percentage loss for the month and the cells are colored to help indicate the performance. Pink indicates a very poor to bad performance or > 5% packet loss. Yellow indicates a poor performance or between 2.5% and 5% packet loss. Green is an acceptable performance or between 1 and 2.5% packet loss, and no background indicates a good performance or < 1% packet loss. These loss thresholds are taken from a Tutorial on Internet Monitoring & PingER at SLAC.
Loss to IN2P3 from PingER monitors

If we break the performance down into regions then we get the graphs below, where the results from the monitoring sites are aggregated by region. The regions are N. America (US and Canada - 14 monitoring sites), Latin America (Brazil), W. Europe (8 monitoring sites in UK, Denmark, Switzerland, Germany, Italy), E. Europe (Hungary), E. Asia (Japan and Korea), Former Soviet Union (Moscow and Novosibirsk), and the Middle East (Israel). The first graph uses a linear y scale, and the second a logarithmiv y scale to better show up the regions with good connectivity. The log scale graph also shows an exponential fit to the median of the all the measurements and appears to indicate that the loss has improved by about a factor of 2 in the last 18 months.

Plans

Gilles Farrache sent email to Les Cottrell November 13, 2000 providing the following planning information for In2P3.

In fact our plans are the following in term of bandwith:

November 2000 : Direct VP ,given by Renater,between ESnet and PHYnet
               (30 Mbps average/ 40 Mbps Peak) given by Renater

Between December 2000
and February 2001 : CERN/US upgrade to 155 Mbps. (no use for US as Lyon/CERN
                    is kept at 34 Mbps)

June 2001 : Lyon/CERN upgrade to 155 Mbps

So in June 2001 we will have several options :

      Trafic to SLAC back to Lyon/CERN link and 
            - Going throught ESnet
            or
            - Going throught Abilene

      Trafic kept throught VP but we do not know what will be the bandwith.

Olivier Martin of CERN provided the following information by email on November 29, 2000:

In less than 2 weeks time we should have STM-1 all the way from CERN, the next 
step could be to turn the protected SONET/SDH circuit into 2 unprotected 
circuits in order to ease experimentation using  4 routers, 2 on each end, one 
for production and another one for research with possibly less stable but more 
advanced software.

We also plan to issue a market survey for so called transatlantic lambda 
service which could be terminated to the new STARTAP (STARLIGHT) but as you 
can guess there are a number of issues around all of this.

Page owner: Les Cottrell