TOCPREVNEXTINDEX

Lund Performance Solutions


Glossary of Terms

CPU Terms

The CPU terms defined in this glossary are specific to the performance data provided by SOS Performance Advisor.
capture ratio
A ratio of time a CPU spent in user mode to system/kernel mode. The capture ratio value is calculated:
Capture Ratio = (User + Real + Nice + NNice) / (Sys + Intr + C SW + Trap + Vflt)
A capture ratio value equal to one or greater indicates the system is spending more than half its time on useful system work. A value of less than one means the system is spending more than half its time on overhead.
context switch
A context switch occurs when a process relinquishes a CPU.
context switch time
The amount of time a CPU spends managing context switches.
high priority time (high pri time)
The amount of time a CPU spends executing high priority processes. A high priority process is any process (excluding batch processes) that does not have a positive nice value. Generally, high priority processes are all interactive and system processes.
idle time
The amount of time a CPU has nothing to do.
interrupt
Interrupt s occur when a high priority event must have control of a CPU. The current running process is forced to temporarily suspend execution while the interrupt is processed. The most well known interrupt is a disk I/O completion interrupt.
interrupt CPU time
The amount of time a CPU spends processing interrupts.
negative nice time (nnice time)
The amount of time a CPU spends in user mode for a process that has a nice level of 0-19. Refer to the nice man page for more information.
nice time
The amount of time a CPU spends in user mode for a process that has a nice level of 21-40. Refer to the nice man page for more information.
real time
The amount of time a CPU spends in in user mode for "real time" priority processes.
system time
The amount of time a CPU spends in kernel mode that does not fall under interrupt, trap, and memory times.
trap
Similar to an interrupt. The difference is that the process currently running on a CPU causes the trap. Interrupts are not caused by the process that is interrupted.
trap time
The amount of time a CPU spends processing traps.
user time
The amount of time a CPU spends in user mode (excluding nice, negative nice, and real times).

Memory Terms

The memory terms defined in this glossary are specific to the performance data provided by SOS Performance Advisor.
activation
An activation occurs when a process is reactivated from a deactivation. See "deactivation".
buffer cache
A pool of buffers in memory with the purpose of maintaining data in memory to avoid disk access.
buffer cache headers
The headers associated with each set of data within the buffer cache.
buffer cache hit
A buffer cache hit occurs when data is found in the buffer cache as opposed to disk. Read hit percentages lower than 90 can indicate the need for a larger buffer cache. Write hit percentage lower than 65 also indicates the potential need to increase the buffer cache size.
deactivation
A deactivation occurs when a process is removed from the list of runable processes because of memory or CPU contention. It will not be scheduled until it is CPU and/or memory contention subsides. Deactivations indicate CPU and/or memory bottlenecks.
desfree
The lower bound for paging. When free memory drops below desfree, paging begins.
dynamic buffer cache (DBC)
The buffer cache is configured in a manner that allows the kernel to dynamically change the buffer cache size. The buffer cache grows as a result of page faults. It shrinks as the vhand process finds unused pages.
fixed size buffer cache
The "fixed size buffer cache" means the size is fixed and will not change without a reconfiguration and recompilation of the kernel.
lotsfree
The upper bound for paging. Once paging has begun, it will continue until free memory is larger than lotsfree.
major page fault
Page faults that require disk access.
minfree
The threshold at which the system considers itself "out of memory." At this point, the system will start deactivating processes.
minor page fault
Page faults that are satisfied in memory; for example, via page reclaims.
page fault
Page faults occur when a page is not found in the buffer cache; the pages are satisfied in memory and disk.
page in
A page in is a page fault that requires disk access.
page out
A page out occurs when the amount of memory required is greater than the amount available. Data within the page is written to disk and the page is made available for use. Excessive page outs indicates a memory bottleneck.
page reclaim
A page reclaim occurs when a requested page exists on the free list. A page reclaim results in a page fault being satisfied in memory.
page scan
A page scan occurs when the vhand process searches through used pages for candidates to page out. Excessive page scanning can be an indicator of a memory bottleneck.
unlockable memory
The amount of memory that cannot be locked. Physical memory that may be locked is called "lockable memory." Locked memory holds frequently-accessed programs or data structures, such as the operating system code. Lockable memory is never more than 3/4 of the available memory. Allowing too much locked memory could lead to a system deadlock. Unlockable memory is used for swapping pages, but lockable memory cannot be used for swapping pages.
VM I/O
A physical disk I/O that is a result of virtual memory management.

Disk Terms

The disk terms defined in this glossary are specific to the performance data provided by SOS Performance Advisor.
logical I/O
An I/O that is satisfied in memory or disk.
physical I/O
An I/O that requires disk access. Physical I/Os include User, Sys, VM, and RAW.
raw I/O
A disk I/O that bypasses the buffer cache.
service time
The amount of time an I/O request takes to be serviced once it begins to be processed by the disk (removed from the disk queue), excluding wait time.
system I/O
A disk I/O that is the result of system overhead in managing files (i.e., super-block reads/writes).
user I/O
A disk I/O that is a result of user file reads/writes.
virtual memory I/O
A disk I/O that is a result of virtual memory management.
wait time
The amount of time an I/O request waits in the disk queue before being serviced. Excessive wait times indicate a disk bottleneck.

Network Terms

The network terms defined in this glossary are specific to the performance data provided by SOS Performance Advisor.
badxid
A duplicate transmission. Every outgoing NFS request is assigned a unique sequential identifier. Requests are retransmitted if the server does not respond within a time-out period. When the server eventually responds, it is possible to respond to the same request multiple times. This is counted as a badxid. badxids are an indication that the server is not responding quickly enough.
collision
A network collision occurs when the system sends a packet at the same time as another system. When collisions occur, the system dispatching them waits a random amount of time to retransmit the packet. Excessive collision percentages indicate a network bottleneck.

Process Terms

The process terms defined in this glossary are specific to the performance data provided by SOS Performance Advisor.
priority
The CPU scheduling priority of the process. High priority numbers indicate low priority status, and vice versa.
think time
The amount of time a process is waiting for user input.
timeslice
The maximum amount of time one process is allowed to run before the scheduler searches for other higher priority processes. The process may give up the CPU sooner if it enters kernel mode.
transactions
A character read or write, or a process death.
wait state
Identifies a resource that a process is waiting (blocked) on.

Wait State Codes

The wait state codes defined in this glossary are specific to the performance data provided by SOS Performance Advisor.
CPU
Executing on the CPU.
JOB
Waiting for job control or tracing signals.
OTHR
Waiting on all other resources and events.
PRI
Waiting for the CPU.
TFLT
Waiting on text page faults.
DFLT
Waiting on data page faults.
KFLT
Waiting on kernel page faults.
ULCK
Waiting on user locks.

Lund Performance Solutions
www.lund.com
Voice: (541) 812-7600
Fax: (541) 81207611
info@lund.com
TOCPREVNEXTINDEX