Wednesday, November 15, 2006

Status Report for Nov 9th to Nov 15th

1. for the perfermance test of Sensorgrid(status: finished)

test 4– Multiple Broker Test (status: finished)
You can find the test structure in the previous report.
Noticed:
We used two benchmark clients to record the time stamp.
Benchmark client 1 connected with Narada broker server1.
Benchmark client 2 connected with Narada broker server2.
The result benchmark client 1 recorded is picuture 1.
The result benchmark client 1 recorded is picuture 2.
Follow the pictures both 1 and 2:
Before the 1th hour, I add 1400 subscribers to system each 10 seconds.After the 1th hour the system works with 1400 subscribers for another 24 hours. The blue line is the 1th hour boundary.
picture 1
picture 2
This test shows that the system with multi-broker scales up to 1400 clients with an acceptable transfer delay. The delay time between two brokers is quite small, just 5 ms.
This test also shows our framework can exceed 1000 subscribers with multi-broker support.
If we want to support huge number clients , we just need add more brokers to a distribute network.

Note:
If NB received this error "TCPReceiverThread: MAJOR ERROR:
Received Data with garbled transport header", the NB will be down.
That error is definitely fatal,maybe it caused by the I/O, since 700 readers
are trying to read the same file, it is possible a garbled data.
I also got some failed tests,maybe those failed test is crushed by server errors or I/O errors.
But we know the minimal thing is that the NB server will be down when
it received this error "TCPReceiverThread: MAJOR ERROR:Received Data
with garbled transport header".
If NB can handle those errors, our sensorgrid framework will be more stable.

The tests enviroment is here:
The perfermance test of Sensorgrid.pdf

The test 1 result is here:
test1_result.tar.gz

The test 2 result is here:
test2_result.tar.gz

The test 3 result is here:
test3_result1.tar.gz
test3_result2.tar.gz

The test 4 result is here:
test4_result.tar.gz

By those 4 tests, we can get:
Our framework is quite strong with these characteristics as follows:
1- stable for continuous operation
2- Has ability to support multiple publishers
3- Has ability to support multiple clients
4- The message delivery times is quite small
5- Has ability to preserve the order of incoming messages


Future work:
GeoFEST stuff(discuss it with Marlon and Yili this Friday )
STfilter portlets and services(discuss it with Marlon this Friday)


Comments: Post a Comment



<< Home