gecko-dev/mstone/ChangeLog.1

130 строки
4.9 KiB
Groff

MailStone 4.1: Changes to since version 4.0
* threads are used on all platforms reduces memory requirements by
10x for large numbers of clients must run as root for maximum
connections per process
* massive internal code and script cleanup / rewrite performance
improvements for socket handling (blocking) and caching files and
name lookups
* filename and path conventions restructuring '/mailstone'
hardwired path no longer required mailstone/results/index.html
(index of test runs) mailstone/results/<datestamp>/stderr (look
here for errors) mailstone/results/<datestamp>/results.html
mailstone/results/<datestamp>/results.txt
mailstone/results/<datestamp>/work.wld (workload config used)
mailstone/results/<datestamp>/testbed.tbd (testbed used)
mailstone/results/<datestamp>/summary.csv (per client summary)
mailstone/results/<datestamp>/*.gif (graphs)
mailstone/tmp/<datestamp>/*.out (raw results from each client)
mailstone/conf/*.wld (workload file) mailstone/conf/*.html
(machine descriptions for reports) mailstone/data/*.msg (sample
test messages)
* periodic statistics reporting allows for trend analysis of many
variables per protocol. sampling rate is automatically determined
by mailmaster Can now generate on-the-fly reports for above by
running "process"
* The accountFormat directive is now obsolete. Use loginFormat
and addressFormat instead.
* The numAccounts and beginAccounts directives are now obsolete.
Use numLogins, firstLogin, numAddresses, and firstAddress instead.
* The sequentialLogins directive disables random account number
selection. This insures a predictable account sequence.
* The checkMailInterval directive for IMAP is now obsolete. Use
loopDelay instead.
* The directives idleTime, loopDelay, and numLoops now apply to
all protocols. See the manual for how loopDelay and numLoops are
used by each protocol.
* a command directive such as <SMTP> without HOSTS=xxx will now
apply to all clients in the testbed
* <include> directive for workload and testbed files (e.g. for
user profile)
* workloads are now passed to 'mailclient' through stdin no test
specific files need to be copied to client machines more
synchonized test startup
* 'setup' script will copy mailclient and test messages
(data/*.msg) to each testbed machine in /var/tmp
* 'cleanup' form of setup will remove mailclient and test messages
from each testbed machine in /var/tmp
* 'checktime' form of setup will (nearly) simultaneously retrieve
time from each client. This lets you easily check for clock
problems.
* 'timesync' form of setup will (nearly) simultaneously set time
on each client. This only works on OSs that support setting
seconds through the "date" command. rdate or ntpdate should be
used if available. You must be the root user and be able to rsh
as root for timesync to work.
* Improved reports in text and html (formatting and content)
* The text version of the report is now only displayed
automatically if "-v" is given.
* Graphs with more than one protocol will also display a "Total"
graph
* The graphs can now be customized for each test (see sample.wld
for the default)
* You can now add graphs (in addition to the default or configured
ones) by using "process timestamp -a conf/moregraph.wld".
* An informative index of test runs is now generated in
results/index.html
* The index is updated while the run is in progress to make
checking for errors easier.
* The error log now displays times relative to test start.
* Memory use while processing results has been greatly reduced.
* A summary of the data from each process is now saved in a
Comma-separated-value (CSV) file. results/timestamp/clients.csv
* A summary of the data over time is now saved in a
Comma-separated-value (CSV) file.
results/timestamp/protocol-time.csv
* new gnuplot binary included, can output directly to gif format
* read and write performance numbers now reported separately
* new runtime banners and copyrights (ooh boy)
* idleTime option for IMAP fixed, plus now applies to all commands
* blockTime now sets a delay between command blocks
* Process model is as follows
User runs the top-level 'mstone' script
This calls 'conf/testname.pl' with command line arguments
This calls 'mailmaster.pl' with config and command line arguments
mailmaster uses 'rsh' or equivalent to launch mailclients
mailclient runs on each testbed and forks numprocs
each proc spawns numthreads (one per client)
<run tests> (note: results are being sent continuously)
threads send results back through sockets to top mailclient
mailclient forwards results back over stdout via rsh pipe
(for future work - do some data reduction/combining here)
mailmaster directs output to each client-<hostname> file
mailmaster launches report generator
* sample LDIF fragment for a postmaster entry included