GCP Coverage Statistic

Additional information for Basket Status page

Date: Wed, 11 Nov 1998 21:54:52 +0100
From: John Walker 
Subject: Re: ** leyline egg

[One of the eggs, leyline, showed coverage statistics at variance from the
others.  The following is John Walker's explanation of what the
statistic means.  There is also a short comment on what version 4 of the
software does, and how the unreleased version 5 differs.]

--------

"Coverage" is calculated by dividing the number of seconds
since the basket was started into the number of valid
(non-missed) samples received from the egg during that period.

This is a real time statistic, so a dial and drop egg's
coverage will fall between connections to the network,
then catch up each time it connects and updates the basket.
Similarly, if an egg continues to run but suffers a
network connectivity interruption, its coverage will
fall until connectivity is restored, then catch up
when samples buffered locally at the egg are delivered
to the basket.

Missed trials count the number of times a scheduled
sample could not be collected in the one-second
interval in which it was intended to be taken.  This
usually happens when the egg host is heavily loaded and
the eggsh program doesn't receive enough CPU time to
complete the sample.  Now the important thing to
note is that missed trials count only trials
*within packets received from the egg* which were
flagged as missed.  If a an entire packet's worth
of samples was not collected (for example, because the
egg host was rebooting at the time or the eggsh program
wasn't running at the time), then the basket will not
count any trials missing because no samples were
received at all.  Whole missing packets will thus reduce
coverage, but not increase the number of trials reported
missing.

These statistics may not be the most transparent way
to present the network status.  I chose them because they
orthogonally distinguish egg uptime and network connectivity
(coverage) and sampling latency problems (missed trials).
As the network matures, we'll probably have to develop
more and better measures of network performance.

>> There is a
>> version 5, but John has never announced it is the working release.

Version 4 was released on September 20, so any egg started
before that is running an older version.  Most of the changes
in Version 4 affected only the basket, and should be
completely compatible with earlier versions of the egg
software.

Version 5 consists at this time of a few cosmetic changes such
as deleting the long-disabled byte-order-sensitive network
and file I/O code, and laying the groundwork for possible
future addition of support for eggs which report via
E-mail (but no actual code for such a facility).  It should
work precisely the same as Version 4, but there is no reason
whatsoever to install it.

Starting with Version 4, the egg and basket software both report
the version number on standard error when started.  Any
version which doesn't is, therefore, prior to Version 4.

GCP Home