- This topic has 3 replies, 2 voices, and was last updated 13 years, 1 month ago by
.
-
Topic
-
Hi, I am using QDX5.7 on Linux.
Scenario: Load testing the thread
I am getting a latency in the translate.
Inbound thread reads a bunch of messages translates them sends them to the server, only after that it reads some more messages. The problem is there is some latency happening in reading the messages.
Is there a way to configure Cloverleaf thread not to wait for the translation . Process continuously without any latency?
Thanks,
Jay
-
Replies
-
-
Jay,
Is your inbound thread a fileset/local protocol?
If so adjust the Scjeduling parameters.
The Scan Interval is how often the directory will be scanned for files – if you only have one file this won’t come into play after the initial scan. If you have one file per message (for example) then lowering this value will lesten the amount of time Cloverleaf will wait to find the next file/mesage.
The Read interval is how long to wait before reading the file (in seconds) that is currently being processed. This wait takes place after the Max. Mesages amount has been read.
Max. Messages is how many messages should be read from the currently processed file once the read interval has been reached.
So if you say read a maximum of 10 records after 3 seconds between reads that means you will get 10 messages every 3 seconds or so. You can alter these values to try to pace the arrival rate.
However if you get messages coming in too fast you will see the route arrows become saturated ant the outbound thread as well – thus not really pacing but rather dumping as fast as possible.
We do that with new integrations to just try and find out the upper saturation level of the outbound thread. Then we throttle back to approximate our required delivery rate to confirm the system can keep up with our minimum service level (3 messages per second).
email: jim.kosloskey@jim-kosloskey.com 29+ years Cloverleaf, 59 years IT - old fart.
-
-
Jay,
You are using PDL TCP/IP on your inbound thread?
If that is the case, then the pacing is really affected by how fast you acknowledge the inbound mesage and how fast the sending system can send the next message.
If you have a batch of message arrive at a time followed by a wait then it is likely the sending system either does not have more messages to send or has some sort of control going to only send x number of messages or only send for x amount of time then cycle again.
When we want to load test an integration we use a fileset/local inbound and control the arrival rate with the scheduling controls I mentioned.
We just resend a bunch of messages from the inbound SMAT file to a file in the directory the Fileset/Local references and then observe the effect of the setting we placed. We adjust the settings to get as near ‘real world’ as possible
email: jim.kosloskey@jim-kosloskey.com 29+ years Cloverleaf, 59 years IT - old fart.
-
- The forum ‘Cloverleaf’ is closed to new topics and replies.
