A simple methodology on Unix systems to approach a problem scenario:
On UNIX, problem investigation often starts with the following:
· Check vmstat or iostat output for the amount of system and user CPU being consumed
versus wait and idle time. Wait time usually, but not always, means waiting on I/O. If
system CPU is equal to or greater than user CPU, this almost always indicates a problem.
· Check the top CPU-using processes by ordering ps output by descending CPU penalty (C)
and memory usage (RSS and SZ columns). For example:
ps -elf | sort +5 -rn
· Are the same processes near the top? Is the CPU penalty staying in the 60-120 range? This
would mean these processes are constantly consuming CPU cycles and need to be
investigated.
On Windows, check for the top CPU and memory users by ordering the output on those columns.
I/O is more difficult to judge without getting perfmon output.
On UNIX, problem investigation often starts with the following:
· Check vmstat or iostat output for the amount of system and user CPU being consumed
versus wait and idle time. Wait time usually, but not always, means waiting on I/O. If
system CPU is equal to or greater than user CPU, this almost always indicates a problem.
· Check the top CPU-using processes by ordering ps output by descending CPU penalty (C)
and memory usage (RSS and SZ columns). For example:
ps -elf | sort +5 -rn
· Are the same processes near the top? Is the CPU penalty staying in the 60-120 range? This
would mean these processes are constantly consuming CPU cycles and need to be
investigated.
On Windows, check for the top CPU and memory users by ordering the output on those columns.
I/O is more difficult to judge without getting perfmon output.
No comments:
Post a Comment