SLAC logo

SLAC to CERN RTT Anomalies Network logo

Les Cottrell. Page created: May 23 2003.

Central Computer Access | Computer Networking | Network Group | ICFA-NTF Monitoring
SLAC Welcome
Highlighted Home
Detailed Home
Search
Phonebook

Introduction

This case shows how the round trip times (RTTs) from two sites (one a users home with a DSL line in Palo Alto and the other SLAC) within a few miles of one another, to CERN in Geneva, Switzerlans could differ by 40 msec (with the DSL line reporting the shorter RTT).

Problem report

Traceroutes indicated that the RTT from Paul Kunz's home DSL to CERN (221ms) is almost 40 ms faster than from SLAC to CERN (258ms).

Understanding the problem

Looking back at the PingER RTTs measured from SLAC to CERN it appears the RTT increased from about 180ms to 240ms around 14:00 on May 22nd 2003 GMT. Looking back for the last year it appears the RTT changes in value several times.

Looking in more detail at the traceroute, it appears the lower latency path is via link to Deutsche Telekom at PAIX. This is confirmed by a traceroute from an Internet2 router in Sunnyvale having pretty much identical latency to the ESnet path.

Apparently Stanford has a direct connection to PAIX in Palo Alto, where they peer with Deutsch Telekom/DTAG (and likely others as well). DT apparently has a direct link between PAIX and their router in Geneva (CIXP-gw20.GVA.CH.net.DTAG.DE). It is not clear how this connection is routed, but it shows a round-trip latency of well under 170msec (probably close to 150msec), which is amazing for a link from the Bay Area to Geneva. Note by contrast, that ESnet is using the CERN link between Chicago (STARLIGHT) and Geneva and it shows a RT latency of well over 200msec.

So when we route through Stanford we are using this very low-latency DT link directly to Geneva and via ESnet (and I2) you will use the longer latency CERN link through Chicago.

In principle ESnet could turn up peering with DT in PAIX and prefer it over the connection through STARLIGHT as the preferred route to CERN, but that seems to have other drawbacks. One possible drawback is that from the PingER measurements the path from SLAC to CERN appears to have a generally higher packet loss.


Page owner: Les Cottrell