HP NonStop G-Series - Manual (Page 156)

Balancing and Tuning a System
Measure User’s Guide — 520560-003
7- 20
Checking and Tuning Problem Areas
Continue the move until the CPU activity at each priority level is balanced across all
CPUs (or is as balanced as it can be given your workload). Move one process at a
time. After moving a process, check the effect of the change. If you shifted the problem
from one CPU to the other, move the process back and try a different process.
The priorities assigned to various types of processes can also affect performance. For
example, it is common to prevent requester processes from interrupting server
processes by keeping the requester processes at a lower priority.
This is one common scheme for setting priorities for processes. The priorities are kept
10 points apart for clarity and to avoid problems. The system bumps a process priority
each time another process is merged in front of another process on the ready list to
ensure that the process is eventually serviced by the disk process.
1. Disk processes (220)
2. Communication processes (200)
3. One command interpreter for problems (190)
4. PATHMON (180)
5. Server processes (170)
6. Requester processes (160)
7. Command interpreters (150)
8. Batch jobs (130)
If possible, separate batch and transaction processing.
Limiting the Number of Processes Measured
The operating system has an architectural limit of 65,534 concurrent processes per
CPU. The actual number of concurrent processes possible in a CPU depends on the
system’s resources, such as memory.
To improve Measure performance and keep the amount of data manageable, set up
the Measure configuration to collect measurements for specific processes rather than
all processes.
Evaluating Response Time
Response time is the user’s chief indication of performance. However, the
RESPONSE-TIME counter for the TERMINAL entity does not measure the response
time as seen by the user. The user sees response time as the time between pressing
the RETURN or function key and the time the system displays a response on the
screen. For Measure, response time is the time between the terminal process receiving
the RETURN or function key and the terminal process posting a write to the terminal.
Note. There are exceptions to the priority levels of requester and server processes. Determine
these case by case by experimentation.