IEPM

TCP Stacks on production links,
a Scalable TCP single stream vs.
Reno TCP multiple streams

SLAC Home Page
Main Page of TCP Stacks on production links | High Speed TCP single streams versus a Reno TCP single stream | Scalable TCP single streams versus a Reno TCP single stream | Fast TCP single stream versus a Reno TCP single stream  | Fast TCP single stream versus Reno TCP multiple streams | Bulk throughput measurements
Fabrizio Coccetti and Les Cottrell. Created 31 March '03, last update 31 March '03

 


Overview | Scalable TCP versus Reno TCP from SLAC to CERN

Overview

In this page we run tests where one new fast TCP stack single stream competes versus multiple Reno TCP streams.
Results shown that the the throughput you can achieve with multiple (2 or more) Reno TCP streams is almost always higher than the throughput you can achieve with one fast TCP stack.

An Iperf stream is started from S1 to the Remote Host.
At the same time, multiple streams are started from S2 to the Remote Host. Adding the throughputs of each of these streams, we get the total throughput from S2 to the remote host.

Scalable TCP versus Reno TCP from SLAC to CERN

In all tests we use MTU=1500B, txqueuelen=100 packets, window size = 8 MB.  Each stream runs for 60 seconds.

The following plot shows, at different times of the day, how a Scalable TCP single stream compares with a competing Reno TCP single stream. The thick vertical black lines separate each set of competing stream measures.  Results show that Scalable TCP outperforms Reno TCP.   To read more about competing single streams with different TCP stacks, you can follow this link.

All the following plots displays, at different times of the day, how a single Scalable TCP stream performs against Reno TCP multiple (from 2 to 24)  streams.