190 строки
5.4 KiB
Plaintext
190 строки
5.4 KiB
Plaintext
Mstone TODO List
|
|
|
|
Updated:
|
|
3-20-2000 Dan Christian
|
|
|
|
======================================================================
|
|
Minor improvements
|
|
|
|
Data reduction at each testbed client
|
|
|
|
Test message generator (plain and MIME styles)
|
|
|
|
More graphs for multiple runs
|
|
|
|
Option to drop a fraction of connections
|
|
|
|
IMAP delete without reading
|
|
Just get the message list and delete everything. This could
|
|
be the fastest way to drain out the store.
|
|
|
|
|
|
Display MIN/MAX or standard deviation on graphs
|
|
Gnuplot can do data points with error bars. You could either
|
|
use MIN/MAX or the standard deviation for the error bars. There are
|
|
issues with calculating the standard deviation numbers throughout the
|
|
graph that need to be addressed.
|
|
|
|
|
|
Statistics reset
|
|
At least the MIN and MAX statistics could be easily reset
|
|
during the test (after ramp up). This keeps the transients during
|
|
startup and shutdown from dominating these numbers. The standard
|
|
deviation statistics are much trickier to reset during the run. It
|
|
may be better to isolate sections in post processing.
|
|
|
|
|
|
Perl web server
|
|
Sometimes it would be nice to include our own web server to
|
|
provide the results (instead of using file: URLs). This would also be
|
|
a gateway to form based test configuration editing and allow results
|
|
to be interactively updated during tests. Perl with a socket library
|
|
could handle this without too much trouble.
|
|
|
|
|
|
Dynamic test loading
|
|
Finalize an API for dynamically loading tests. This would
|
|
allow tests to be added or updated separately from the core
|
|
functionality. This may be needed for some types of security testing.
|
|
|
|
|
|
Link graphs to the results tables
|
|
There are already tags by each graph. You should be able to
|
|
link the appropriate results table entry to each graph. This sort of
|
|
tricky since graphs often combine multiple entries.
|
|
|
|
|
|
Show statistics for every block
|
|
Statistics are actually kept for every block in every thread.
|
|
There should be a way to view information at this detail.
|
|
|
|
|
|
Man pages
|
|
The online docs are nice, but good 'ol man pages would be
|
|
great in a different way.
|
|
|
|
|
|
Reduce namespace polution
|
|
Scripts names like setup and cleanup are too general. They
|
|
|
|
should be part of the main 'mstone' script (e.g. mstone setup).
|
|
|
|
Examples of script series
|
|
Include example scripts to run entire series of tests in a
|
|
sane way.
|
|
|
|
|
|
Fix FORMAT clash
|
|
At the start of a test, each client process outputs the
|
|
information needed to report all its protocols. When there are
|
|
multiple processes on one client, these FORMAT lines can intermix and
|
|
cause parsing errors.
|
|
|
|
|
|
Set connection drop rate
|
|
Drop some percentage of the connections without a proper
|
|
shutdown. This tests how well a server can detect and recover from
|
|
hard disconnects.
|
|
|
|
|
|
Improve randomness
|
|
The way that we generate random numbers in a range may be not
|
|
generate the proper randomness. We are using lrand48()%range we
|
|
should use (lrand48/RAND_RANGE)*range. There are some end conditions
|
|
that need to be thought about. All of this is in the sequence code;
|
|
one change should fix (or break :) everything.
|
|
|
|
Also, we may be generating numbers that are never used. This
|
|
may be costly, and can create holes in the proper sequence.
|
|
|
|
|
|
Improve printing
|
|
The color graphs are great on screen, but tend to print
|
|
poorly. Either we need a better way to generate a printable version
|
|
(maybe through a CGI button), or the seperate protocols need to be
|
|
printed individually. Also, Communicator does a lousy job of keeping
|
|
title with tables or graphs. Hopefully, Mozilla will do better.
|
|
|
|
|
|
======================================================================
|
|
Whole new protocol tests:
|
|
|
|
ICQ
|
|
Test high volume instant messaging. Very useful for the
|
|
bridges and gateways that people are considering using.
|
|
|
|
|
|
WAP
|
|
WAP is the emerging standard for mobile phones.
|
|
|
|
|
|
WCAP
|
|
Web based calendar services
|
|
|
|
|
|
LDAP
|
|
Use the LDAP SDK to do basic LDAP testing. The SDK probably
|
|
isn't fast enough to call this a real performance test, but you can at
|
|
least test performance degredation due to load from a real application
|
|
(like a mail server).
|
|
|
|
|
|
DNS
|
|
Mail servers use DNS a lot. You should at least be able to
|
|
see if performance is degrading due to load.
|
|
|
|
|
|
Disk/filesystem
|
|
Test read, write, sync, link, un-link, and append performance under
|
|
multiple threads.
|
|
|
|
|
|
Cert servers
|
|
Test certificate authenticity checking performance
|
|
|
|
|
|
======================================================================
|
|
Possible dummy servers:
|
|
|
|
SMTP
|
|
Receive mail via SMTP and ignore it. Usefull for SMTP relay
|
|
testing.
|
|
|
|
|
|
DNS
|
|
Simulate slow DNS server lookups. Usefull for SMTP relay testing.
|
|
|
|
|
|
======================================================================
|
|
Major changes
|
|
|
|
Throttling
|
|
Monitor and control transaction rates so that specific load
|
|
levels can be easily specified. The rates should be able to vary to
|
|
simulate peek hour usage and disconnect-restore.
|
|
|
|
|
|
NSPR threading
|
|
Use NSPR for threading and sockets. This may allow other OSes
|
|
to be used as clients. This might be easy, since mstone does not need
|
|
any locking, just simple thread creation and harvesting. NSPR
|
|
argument parsing and hashes may also be useful.
|
|
|
|
|
|
SSL
|
|
Support SSL on the protocols that allow it. May require NSPR.
|
|
|
|
|
|
Line speed emulation
|
|
Simulate the variable delays and limited throughput of dial up
|
|
connections.
|
|
|
|
|
|
Scripting
|
|
Allow more detailed control of protocol tests. It looks
|
|
difficult to make this scalable and fast.
|
|
|
|
|
|
Combined tests
|
|
Deliver mail over SMTP and then see when it arrives using IMAP.
|