Performance Metrics - Development Environment vs PT Environment
Test 1
2 copies of HL7Spy, one configured to listen on port 12000, one set to send on port 12000. Chicago data set (100K messages)
Send Machine | Receive Machine | Time to send 100K | Notes |
|---|---|---|---|
JR-PC | JR-PC | 15 sec | 4 Cores (8 Hyperthreaded Cores) |
NICKWORK-PC | NICKWORK-PC | 12 sec | 4 Cores (8 Hyperthreaded Cores) |
1st Gen I3 computer | 1st Gen I3 computer | 23 sec | 2 Cores |
PT Load Generator | PT Load Generator | 57 sec | 2 VM Cores |
PT Application Server 02 | PT Application Server 02 | 75 sec | 4 VM Cores |
PT Database Server 01 | PT Database Server 01 | 78 sec | 6 VM Cores |
JR-PC | NICKWORK-PC | 36 sec |
|
PT Load Generator | PT Application Server 02 | 127 sec | very jerky performance. At times no messages are sent for a period of 10 seconds or more |
PT Load Generator | PT Database Server 01 | 162 sec | very jerky performance. At times no messages are sent for a period of 10 seconds or more |
PT Application Server 02 | PT Database Server 01 | 185 sec | very jerky performance. At times no messages are sent for a period of 10 seconds or more |
Jonathan Reis' presentation on the PT environment
Observations
We are expecting the best performance to be between the Application Servers and the Database Server since the Application Server must round-trip to the database for every messaging operation in Connexion. Yet we are seeing the worst performance between these 2 machines.
Counter-intuitively, the more cores we have the slower the performance
There seems to be an issue with the writing out to disk. Long pauses cause the applications to freeze intermittently and sometimes timeout.
In our local network we can copy a 90MB file in about 1 second. In the PT environment the time varied between 8 and 20 seconds with very jerky progress.
Here is a video of what it looks like to copy a 90MB file between 2 systems in the PT environment https://dl.dropboxusercontent.com/u/83119949/WP_20140217_008.mp4