NAME
pmshub - tool to monitor SN2 cache traffic and CPU uti
lization
SYNOPSIS
pmshub [-A] [pmgadgets option ...]
DESCRIPTION
pmshub is a tool for monitoring cache traffic on SGI SN2
systems. It will not work on any other platform.
The command line options are:
-A Activate SHub cache traffic performance monitoring
on the target host, and then exit. The target host
is specified with the -h option to pmgadgets(1).
-V Print the generated configuration file on standard
output before running pmgadgets. Brave individuals
may use this to customize the configuration file.
-t interval
Specify the sample interval for updating metrics.
For hosts on slow or congested networks, it is
advisable to use a larger interval. The default is
2 seconds.
See the man page for pmgadgets(1) for other applicable
options.
The pmshub display is divided into cells, one for each
node in the system. On the left hand side, the cell for
each node shows the logical node number and it's geograph
ical identity (rack, slot, slab and NASID);
On the top right of each node cell a scrolling CPU uti
lization chart for each CPU on the node is displayed.
These charts show "user" CPU utilization in dark blue,
"nice" in cyan and "system" in red.
The three strip charts across the bottom of each node cell
show cache traffic as seen by the SHub memory controller
on each SN2 node board. Each node board has it's own
unique range of physical addreses and a load or store to
any particular address may involve the SHub, cache and
main memory on a remote node, but the reference will be
counted as either local or remote for the originating
node. The scale ticks on the left side of each chart
range from 0 to 8 million cache operations per second.
Since each SN2 cacheline is 128 bytes, this can roughly be
translated into memory bandwidth for references that miss
in the cache. It important to note however that none of
either load or store operations.
Each strip chart shows green for local cache traffic and
yellow for (more expensive) remote off-node traffic. The
three charts have the following interpretation:
Read Miss - Data
The read data miss rate - this indicates the
rate of data load operations for addresses that
are not currently cached. The instruction miss
rate metric is available, but is not displayed
by pmshub.
Write Line
The write back rate - this indicates the rate at
which evicted dirty cachelines ("victims") are
written back to main memory.
Invalidates
The read/invalidate line rate - this indicates
the rate at which cachelines are being invali
dated due to cacheline flushes or store opera
tions (which occur either locally or on another
node).
Pressing the middle mouse button over a gadget inside the
pmshub window (or pressing the space bar with the mouse
pointer over a gadget) brings up the "Info" dialog which
shows the current value (as text) for the metric(s) dis
played by the gadget under the mouse pointer. Pressing
the right mouse button inside the pmshub window brings up
a popup menu which allows one to terminate pmshub. Press
ing the escape key (Esc) with the mouse pointer in the
window will also terminate the application. At this time
there are no other "actions" that can be launched from
pmshub.
FILES
$PCP_VAR_DIR/pmns/*
default PMNS specification files
PCP ENVIRONMENT
Environment variables with the prefix PCP_ are used to
parameterize the file and directory names used by PCP. On
each installation, the file /etc/pcp.conf contains the
local values for these variables. The $PCP_CONF variable
may be used to specify an alternative configuration file,
as described in pcp.conf(4).
SEE ALSO
pmgadgets(1) and pmgsys(1).
Man(1) output converted with
man2html