HP-UX TCP/IP Performance White Paper, March 2008

38
aspect of paging is that it can cause external fragmentation of the memory. This fragmentation will cause
fewer large pages available for use by applications.
6.2.4 Monitoring Memory utilization using vmstat
The vmstat(1)command reports certain statistics kept about process, virtual memory, trap, and CPU
activity. For example:
# vmstat -S
procs memory page faults
cpu
r b w avm free si so pi po fr de sr in sy cs us
sy
1 0 0 71646 758881 0 0 0 0 0 0 0 2467 21014 360 1
3
The key vmstat performance indicators are the following:
• pi page in
• po page out
6.2.5 Monitoring Cache Miss Latency
The processor caches used for data and instructions can greatly affect the performance of the system.
Caches are used specifically to store information that is read in with the expectation that it will be needed
again soon by the CPU. In a multitasking environment, the cache contains entries belonging to many
different processes, in the expectation that the processes will execute again before needed locations are
flushed by other processes.
Cache problems become visible as overuse of user or system CPU resources. You can see this in CPU
utilization metrics, or you can compare performance metrics on systems with different cache sizes or
organizations. Unfortunately, it is very difficult to detect cache problems directly. One system with millions
of cache misses per second may actually be performing well while another system with several hundred
thousand a second may not perform well. What really matters is how well the processor is able to overlap
the cache miss time with useful work.
Caliper can be used to monitor the cache misses. In addition, other cache metrics such as latency per miss
can be tracked. It is a very useful tool for analyzing cache behavior on HP Integrity Servers. The "dcache"
option in caliper can be used to find data access performance problems.
6.2.6 Monitoring other resources
Monitoring the performance of other resources such as switches and I/O backplane bandwidth, which can
potentially become bottlenecks for network performance, is also essential for improving network
performance.
6.3 Measuring Network Throughput
The netperf utility is a very useful tool to measure the throughput of the network and various other
aspects of network performance. It can generate traffic patterns such as bulk data transfer and
request/response type of data transfer. It is also very useful for measuring performance for TCP, UDP, and
the datalink layer. The complete reference manual is available at http://www.netperf.org.