Hi, all:
Just a quick note to let you know that a new EROS-related paper,
The Measured Performance of a Fast Local IPC
Is available from the EROS home page via the "Technical Documentation and Papers" link. The EROS home page can be found via
http://www.cis.upenn.edu/~eros
The results were pleasantly surprising.
One note on the numbers in the final chart. In spite of our best attempts to minimize the intrusion of other applications, we found that TCP, UDP, and pipe throughput numbers vary a lot (in some cases by a factor of two). EROS IPC, by contrast, is highly reproducible. We decided to take "best of 3" for TCP, UDP, and pipe measurements, but this is not what you would see in practice.
Also, the number that we think is the most interesting is the bottom line in the bottom left corner of that chart (the one with the triangles). This is the mode used by the RPC library, and the bottom left corner covers the most common payload size. Note that the line shows the TCP transport costs only. If we factored in the RPC library overhead (in retrospect I should have measured and included that), the throughput would be lower.
Jonathan
From shap Tue Nov 5 21:52:30 1996
Return-Path: shap
Received: (from shap@localhost) by eros.cis.upenn.edu (8.7.6/8.7.3) id VAA01244 for eros-announce-test; Tue, 5 Nov 1996 21:52:26 -0500
Date: Tue, 5 Nov 1996 21:52:26 -0500
From: "Jonathan S. Shapiro" <shap>
Message-Id: <199611060252.VAA01244@eros.cis.upenn.edu>
To: eros-announce-test
Subject: test of announce
verifying alias