performance issues? first stop vmstat. vmstat is a very useful tool because allows you to have a quick overall performance view of your system and is available in all Unix type systems here some quick a dirty explanations on how to understand the output of the vmstat command... btw I am working on AIX and we will focus on a simple output 1. r should never be smaller than b or we may have a CPU bottleneck due processes suspended due to memory load control 2. if fre is really small and if any paging is going on pi or po this is most likely a cause of a bottleneck 3. if b and wa are high we may have an I/O bootleneck due the number of blocking processes 4. if b is low or normal and free is small and us + sy = (close to 100) then we have a memory bottleneck 5. if us+sy average more than 80% we may have a CPU bootleneck if you are at 100 our system is breathing heavily 6. if us+sy is small but wa is greather than 25 we may have I/O intensive activitie or disk subsystem might not be balanced properly which turns on cpu not being able to work as hard as he can 7. if us+sy is over 80% and r is larger than [5 * (Number of processors - Number of bound processors)] then we have a CPU bound 8. if r is greater than the number of CPUs, there is at least one thread waiting for a CPU and this is likelihood of a performance impact. 9. if sy raises over 10000 per second per processor we may be polling subroutines likes select() indicates a bad code it is advisable to have a baseline measurement that gives a count for a normal sy value.
performance issues? first stop vmstat.
vmstat is a very useful tool because allows you to have a quick overall performance view of your system and is available in all Unix type systems
here some quick a dirty explanations on how to understand the output of the vmstat command... btw I am working on AIX and we will focus on a simple output
1. r should never be smaller than b or we may have a CPU bottleneck due processes suspended due to memory load control
2. if fre is really small and if any paging is going on pi or po this is most likely a cause of a bottleneck
3. if b and wa are high we may have an I/O bootleneck due the number of blocking processes
4. if b is low or normal and free is small and us + sy = (close to 100) then we have a memory bottleneck
5. if us+sy average more than 80% we may have a CPU bootleneck if you are at 100 our system is breathing heavily
6. if us+sy is small but wa is greather than 25 we may have I/O intensive activitie or disk subsystem might not be balanced properly which turns on cpu not being able to work as hard as he can
7. if us+sy is over 80% and r is larger than [5 * (Number of processors - Number of bound processors)] then we have a CPU bound
8. if r is greater than the number of CPUs, there is at least one thread waiting for a CPU and this is likelihood of a performance impact.
9. if sy raises over 10000 per second per processor we may be polling subroutines likes select() indicates a bad code it is advisable to have a baseline measurement that gives a count for a normal sy value.
here some quick a dirty explanations on how to understand the output of the vmstat command... btw I am working on AIX and we will focus on a simple output
1. r should never be smaller than b or we may have a CPU bottleneck due processes suspended due to memory load control
2. if fre is really small and if any paging is going on pi or po this is most likely a cause of a bottleneck
3. if b and wa are high we may have an I/O bootleneck due the number of blocking processes
4. if b is low or normal and free is small and us + sy = (close to 100) then we have a memory bottleneck
5. if us+sy average more than 80% we may have a CPU bootleneck if you are at 100 our system is breathing heavily
6. if us+sy is small but wa is greather than 25 we may have I/O intensive activitie or disk subsystem might not be balanced properly which turns on cpu not being able to work as hard as he can
7. if us+sy is over 80% and r is larger than [5 * (Number of processors - Number of bound processors)] then we have a CPU bound
8. if r is greater than the number of CPUs, there is at least one thread waiting for a CPU and this is likelihood of a performance impact.
9. if sy raises over 10000 per second per processor we may be polling subroutines likes select() indicates a bad code it is advisable to have a baseline measurement that gives a count for a normal sy value.
From:
0 comments:
Post a Comment