Enterprise Server Kernel tuning
Kernels Tuning:
================
Below are some Oracle recommended tuning parameters.
Call Object Kernel:
This number may vary based on how "heavy" the interactive use is. If there aren't many BSFN calls or the BSFNs are short and quick, then you might be able to get more users per CO kernel. If there are a LOT of BSFN calls, BSFNs that are long-running or customized BSFNS, then you may need fewer users per CO kernel.
Configuring parameter singleThreadedMode=N
Prior to 8.11SP1/TR 8.95 this parameter was set to Y. The COK ratio recommendation is 1 kernel per 6-10 users.
After 8.11SP1/TR 8.95 this parameter should be set to N and the COK ratio recommendation is 1 kernel per 17-20 users.
Setting this parameter to N improves performance.
Note: Call Object Kernel requires testing and tuning.
The recommendations are meant only as a guideline and should be tested thoroughly (customizations, user load, all modules
Multi-Threaded Kernels:
Multi-Threaded Kernels is available starting with EnterpriseOne 8.11 SP1, Multi-threaded kernels can improve performance by allowing the engine to process more user requests simultaneously. Specifically, business functions will be able to run in multiple threads within the kernel processes on the Enterprise server. This eliminates performance bottlenecks within kernel processes and results in more consistent response times for end users.
MaxNetConnections:
The maxNetConnections parameter must be high enough to support all the socket connections to AND from your Enterprise Server. This includes, but is not limited to JAS Servers, Windows clients, Events servers or clients, and XPI.
For example, if you have 10 JAS/HTML Servers, each with a maxPoolSize=50 (in the JAS.INI), and 10 Windows clients that can potentially be connecting at the same time; then the way to calculate MaxNetConnections in this example would be 10 (for the Windows clients, one bidirectional connection for each FAT client) + 10 x 50 (maxPoolSize for each JAS server), resulting in 510 as the value. This does not include any XPI connections or other possible socket connections to/from the Enterprise/Application Server that may exist in your configuration.
maxNetConnections also depends on the maximum number of concurrent EnterpriseOne users that the system is expected to handle, as well as the processing and memory power of the Enterprise Server.
JDENET_N
The ratio for JDENET_N processes to interactive users is 1 jdenet_n process for every 25-50 interactive users. So, for 650 interactive users, our recommendation is to set maxNetProcesses to 14-27 (this is 1 more than 13-26 since the first jdenet_n does not really handle client communication; it does the initial connection routing, outbound connections, and process management). To set maxNetProcesses you must first determine the maximum number of connections that you need and the number of connections a single process can handle.
IPC Setting
To support the above processes, you need to make sure your EnterpriseOne IPC settings are high enough as well. Here is a description of how the E1 processes relate to the IPC settings:
The number of message queues that each instance of E1 will need can be determined from the following:
numQueues=net_n + 2*net_k + ube + misc
net_n is the max number of jdenet_n processes that will run concurrently
net_k is the max number of jdenet_k processes that will run concurrently
ube is the max number of UBEs that will run concurrently
misc is the max number of miscellaneous E1 processes that will run concurrently (porttest, runprint, etc.)
The number of semaphores that each instance of E1 will need can be determined from the following:
numSemaphores=pathcode_factor
pathcode_factor roughly depends on number of pathcodes plus a fixed amount. The recommendation is: 50 + 40 * (num_pathcodes)
The number of shared memory segments (numSharedMemSegs) that each instance of E1 will need is difficult to quantify. There is a fixed number that the instance will need (possibly 20), but then each instance will need a variable amount more which depends on the following:
Number of net and kernel processes
Amount of data being passed from process to process on the server
Size of data being passed from process to process on the server
An E1 instance that has many net and kernel processes passing few small messages between processes might not need any more shared memory segments. Whereas an E1 instance that has few net and kernel processes passing many large messages between processes might need many more shared memory segments.
Refer Oracle documentation for more information..
================
Below are some Oracle recommended tuning parameters.
Call Object Kernel:
This number may vary based on how "heavy" the interactive use is. If there aren't many BSFN calls or the BSFNs are short and quick, then you might be able to get more users per CO kernel. If there are a LOT of BSFN calls, BSFNs that are long-running or customized BSFNS, then you may need fewer users per CO kernel.
Configuring parameter singleThreadedMode=N
Prior to 8.11SP1/TR 8.95 this parameter was set to Y. The COK ratio recommendation is 1 kernel per 6-10 users.
After 8.11SP1/TR 8.95 this parameter should be set to N and the COK ratio recommendation is 1 kernel per 17-20 users.
Setting this parameter to N improves performance.
Note: Call Object Kernel requires testing and tuning.
The recommendations are meant only as a guideline and should be tested thoroughly (customizations, user load, all modules
Multi-Threaded Kernels:
Multi-Threaded Kernels is available starting with EnterpriseOne 8.11 SP1, Multi-threaded kernels can improve performance by allowing the engine to process more user requests simultaneously. Specifically, business functions will be able to run in multiple threads within the kernel processes on the Enterprise server. This eliminates performance bottlenecks within kernel processes and results in more consistent response times for end users.
MaxNetConnections:
The maxNetConnections parameter must be high enough to support all the socket connections to AND from your Enterprise Server. This includes, but is not limited to JAS Servers, Windows clients, Events servers or clients, and XPI.
For example, if you have 10 JAS/HTML Servers, each with a maxPoolSize=50 (in the JAS.INI), and 10 Windows clients that can potentially be connecting at the same time; then the way to calculate MaxNetConnections in this example would be 10 (for the Windows clients, one bidirectional connection for each FAT client) + 10 x 50 (maxPoolSize for each JAS server), resulting in 510 as the value. This does not include any XPI connections or other possible socket connections to/from the Enterprise/Application Server that may exist in your configuration.
maxNetConnections also depends on the maximum number of concurrent EnterpriseOne users that the system is expected to handle, as well as the processing and memory power of the Enterprise Server.
JDENET_N
The ratio for JDENET_N processes to interactive users is 1 jdenet_n process for every 25-50 interactive users. So, for 650 interactive users, our recommendation is to set maxNetProcesses to 14-27 (this is 1 more than 13-26 since the first jdenet_n does not really handle client communication; it does the initial connection routing, outbound connections, and process management). To set maxNetProcesses you must first determine the maximum number of connections that you need and the number of connections a single process can handle.
IPC Setting
To support the above processes, you need to make sure your EnterpriseOne IPC settings are high enough as well. Here is a description of how the E1 processes relate to the IPC settings:
The number of message queues that each instance of E1 will need can be determined from the following:
numQueues=net_n + 2*net_k + ube + misc
net_n is the max number of jdenet_n processes that will run concurrently
net_k is the max number of jdenet_k processes that will run concurrently
ube is the max number of UBEs that will run concurrently
misc is the max number of miscellaneous E1 processes that will run concurrently (porttest, runprint, etc.)
The number of semaphores that each instance of E1 will need can be determined from the following:
numSemaphores=pathcode_factor
pathcode_factor roughly depends on number of pathcodes plus a fixed amount. The recommendation is: 50 + 40 * (num_pathcodes)
The number of shared memory segments (numSharedMemSegs) that each instance of E1 will need is difficult to quantify. There is a fixed number that the instance will need (possibly 20), but then each instance will need a variable amount more which depends on the following:
Number of net and kernel processes
Amount of data being passed from process to process on the server
Size of data being passed from process to process on the server
An E1 instance that has many net and kernel processes passing few small messages between processes might not need any more shared memory segments. Whereas an E1 instance that has few net and kernel processes passing many large messages between processes might need many more shared memory segments.
Refer Oracle documentation for more information..
Comments
Post a Comment