post production monitoring is always helpful to capture webserver traffic details and provide online traffic details .Best tools are webtrends and hp open view.
The main area need to be focused for performance tuning are CPU,Memory,Disk and Network.
Processor Bottlenecks:
%processor utilization (processor_total: % processor_time) - consistent more than 80% utilization is bottle neck
% user time (processor_total: % user_time) - - this shows application process is taking more utilization
%privileged time (processor_total: % privileged_time) -this shows how much system is busy with system kernel activities
%i/o (in case of unix) - it indicates whether threads are waiting or not, the percentage wait for i/o completion activity .
processor_queue_length - it shows how many threads are waiting in queue,basically more than two indicate queue length is a problem.
interrupts - more than 1000 per second indicates hardware failure or driver config issues
context switches - more than 15000 per second indicates threads are blocked
Memory counters :
available bytes : consistent value less than 20% indicates memory issue
pages per second - it indicates number of pages stored/retrieved from secondary memory due to paging,more than 5 indicates memory problem
page faults per second - how many times information is not available from primary memory ,hard page faults indicates short of memory
working set - this is number of pages that process can address without paging from primary memory
memory page reads
memory page writes
Disk Counters :
%disk utilization : this indicates how much busy is disk
disk_queue_length - number of specified+ 2 indicates threads waiting for disk
avg_disk_read_length
avg_disk_wtire_length
avg_disk_transfer
Logical Disk - Disk reads/sec
Network counters :
Total bytes - this refers total bytes sent and received .At any given point 80% of bandwidth is a prblem
bytes sent
bytes received
No comments:
Post a Comment