Kbase P105094: Why the WebSpeed Agents / AppServer Servers decrease after while.
| Autor |
Progress Software Corporation - Progress |
| Acesso |
Público |
| Publicação |
01/08/2008 |
|
Status: Verified
GOAL:
How does the broker calculate currently numbers of the running Agents and Servers?
GOAL:
Why does WebSpeed Agents / AppServer Servers decrease after while?
GOAL:
After while the number of WebSpeed Agents have decreased.
GOAL:
After while the number of AppServer Servers have decreased.
GOAL:
How does the ubroker.properties autoTrimTimeout feature work?
GOAL:
How does the autoTrimTimeout value for WebSpeed and AppServer works?
GOAL:
How does the WebSpeed Broker spawn and trim Agents?
GOAL:
How does the maxSrvrInstance and the minSrvrInstance work?
GOAL:
How does the Agent/Server Auto-trim value affect the numbers of running Agents/Servers?
GOAL:
How does the Algorithm for the autotrim feature in WebSpeed and AppServer works?
GOAL:
autoTrimTimeout
FACT(s) (Environment):
All Supported Operating Systems
Progress 9.x
OpenEdge 10.x
FIX:
The Agent/Server Auto-trim value.
The 'autoTrimTimeout' value is the time the broker attempts to automatically trim / remove the idle servers/agents for the broker. When the autoTrimTimeout is reached, the broker will trim available server/agents down to the number defined as the minSrvrInstance. BUSY Agents or Servers will never be trimmed.
The lower you set the Agent/Server Auto-trim Interval value, the more quickly the broker reacts to a drop in the workload. Under a reasonably steady work load, the broker should trim the number of running agent/servers to the number needed. However, if the workload fluctuates a great deal and the Agent/Server Auto-trim Interval is set too low, the broker may trim agent/servers that it must then restart.
Note that repetitive trimming and restarting causes the system to waste resources on unnecessary process management. Setting the Auto Trim Timeout to 0 disables it and no trimming will be done.
An Agent/Server Auto-trim Example:
To better understand the use of the Agent/Server Auto-trim feature, consider a scenario involving:
- a broker serving a client/server application running in stateless mode
- following property values are set:
Minimum Agent/Server Instances: 10
Initial Agent/Server Instances to Start: 10
Maximum Agent/Server Instances: 50
Agent/Server Auto-trim Interval: 1800
When the broker starts, it starts 10 agent/servers. This also marks the start of the first Agent/Server Auto-trim Interval. Assume that the broker is idle for 30 minutes (it's just before people arrive for work, for instance). At the end of the 30-minute interval, the highest number of agent/servers that were busy at the same time would be zero (no requests made of any of the agent/servers).
Since zero is less than the Minimum Agent/Servers Instances value of 10, no automatic trimming takes place. This marks the start of the second interval.
Agents 10 10
Time 0----------15----------30---->
Status A A
Trim 0
A = Available
B = Busy
As the second interval starts, employees begin their workday and a sudden need for additional agent/servers occurs. In fact the broker receives so many requests that it must start 20 new agent/servers. Assume this demand on the agent/servers continues through the second interval.
At the interval's end, the maximum number of busy agent/servers would be 30, which is the number the broker scaled up to in response to high demand. 30 agent/servers are running, and the maximum busy level is 30. No trimming occurs.
Agents 10 10 30
Time 0----------30----------60---->
Status A A B
Trim . 0
A = Available
B = Busy
During the next interval, the request load diminishes and at the end only 20 agent/servers are concurrently busy. The maximum number of busy agent/servers for the interval remains at 30, since 30 were busy at the start of the interval. Thirty agent/servers are running, and the maximum busy level is 30. No trimming occurs.
Agents 10 10 30 30
Time 0----------30----------60----------90---->
Status A A B 20B/10A
Trim 0
A = Available
B = Busy
Assume that all employees have to attend a meeting that begins midway through the interval. The highest number of agent/servers busy at that same time was 20.
At the end of the interval, there would be 30 agent/servers running, with zero agent/servers currently busy. Since the maximum busy level is 20 at the start of the interval) while the number of running agent/servers are 30 (all available since everyone in meeting), the broker will trim back the number of agent/servers to 20 (the highest number of busy agent/servers at that same in the interval).
Agents 10 10 30 30 30 ..20
Time 0----------30----------60----------90----------120---->
Status A A B 20B/10A A
Trim 0 10 -> 20 reminding
A = Available
B = Busy
The company meeting ran for another 30 minutes, and the entire next interval elapsed without any requests. Therefore the maximum busy agent/servers would be zero, with 20 currently running. Since the Minimum Agent/Server Instances property value is set to 10, the broker will trim back to 10 agent/servers.
Agents 10 10 30 30 &nb.sp; 30 ..20 20 ..10
Time 0----------30----------60----------90----------120----------150---->
Status A A B 20B/10A A A
Trim 0 10 10 -> 10 reminding
A = Available
B = Busy
.