What's New

April 19, 2004

Latest test results:

New document: Latest Analysis (.XLS) 3.65 MB

April 6, 2004

More performance analysis have been conducted. A summary is in the document below.

New document: Performance Analysis (.XLS)

April 1, 2004

Jitter problem yet remain unsolved! Our initial performance graphs and data show extreme spikes and randomness that
it is hard to analyze the collected data. This could be caused by numerous sources such as network latency,
CPU load on the server, and connection time to another server when primary fails.

Ideas and approaches we are going to take to trace down the jitter problem:

1) First BIG spike. This is significantly steeper than the rest of the data, which distorts the graphs. The best solution is to "temporarily" leave it out of our analysis.
2) Other big spikes. These can be decreased by reducing connection-establishing time. Everytime the primary fails, extra time is required to establish connection to the backups. This time could be reduced by pre-establishing connections with the backups when connecting to primary. This should reduce our spikes and jitter whenever a server fails and switch over to backup.
3) Network. Our jitter could very well caused by network unstability. During busy hours, network latency will increase, hence affect our performance measurement. In order to make sure it is the network, we will test our system using one computer.
4) Extreme rise and falls. This could be elminated, or at least contained by using the Three Sigma approach, which will define a max and min threshold to reduce errors and unstability.
5) CPU load. In addition to taking measurements of our system, we have also been (and will always do) take measurements of the CPU load. It is definitely one source of our jitter problem.

An analysis and summary of test results will be posted shortly.

March 30, 2004

Project baseline -- completed!
Fault Toloerance -- completed!
Performance -- in progress.

Currently taking measurements and improving our system performance through reducing down-time during faults and
possibly load balancing.

Stay tuned!!

New Documents: Client Log -- Fault-Free (.XLS)
  Repman/Fault Tolerance Memo (.DOC)


February 22, 2004

Website is revamped!
New content: Reliability Basic Design (.DOC)