This was a very interesting presentation by a person from CERN. Mainly focused on some nice tricks you might wann know when the systems does not behave like it should with VMM and Memory allocation, sockets and so on.
- th * => actuall processlist
- cpu => pointer to currecnt thread listet in the system
- lrustate <number>
- checks recently used pages within the memory pages
- old pages will be paged out to jfs/jfs2
- paging in general is not bad, only if active app-data are paged in/out every minute
- if a RAC system pages, the system is more or less dead.
- vmstat => info on current memory usage (pinned page size / etc)
latency vs throughput of memory pages
TSM performance read-ahead and write-behind are most important for fast TSM transactions
read-ahead in tunable vmo:
write-behind in tunable vmo:
- svmon -P => call processes with memory usage
- svmon -p PID => info for 1 process
Sockets within AIX and who uses them
Normaly it is not easy to find out who uses a socket within the AIX system. Here is a way, not easy, but works:
netstat -An (find the socket and the Pointer-number within the list)
kdb => sockentinfo POINTERNUMBER
A SLOT value will be shown which is equal to the aix processnumber
A real nice idea is to mount a FS via SSH from a remote system instead of using NFS. With this, all permissions will be mounted also.
sshfs user@server:/directory /mountpoint
fusermount -u /mountpoint