Ganglia
Distributed monitoring
system
-Tirumal
Cluster?
A cluster is a collection of
computers which work together in
accomplishing a task.
Cluster computing has become a
practical choice for high-
performance computing (HPC)
deployment.
1
Ganglia?
Ganglia - A real-time cluster monitoring tool
that collects information from each
computer in the cluster and provides an
interactive way to view the performance of
the computers and cluster as a whole.
Ganglia like other monitoring tools only
provide a way to view but not control the
performance of each computer.
Ganglia Architecture
2
Ganglia –A monitoring tool
Ganglia consists of two parts
gmond (ganglia monitor daemon)
gmetad
Gmond: Runs on every node of the cluster
and collects data about the node like
CPU load, free memory, disk usage,
network traffic, etc.
Gmetad: Runs on a head node, gathers the data
from all the nodes, and displays it.
Ganglia is scalable as we can gather other
metrics of interest, send them to the host and
display them.
It is currently in use on over 500 clusters
around the world, can handle clusters with 2000
nodes.
3
A snapshot of our enhanced
Ganglia
Adding Metrics to Ganglia
Modifying the source code
Using the gmetric tool (provided by Ganglia)
4
Modifying the source code
The Ganglia source code includes three
files specific to metrics.
/gmond/key_metrics.h
/gmond/metric.h
/gmond/machines/linux.c
Key_metrics.h
enum {
cpu_num,
cpu_num,
cpu_speed,
cpu_speed,
mem_total,
mem_total,
swap_total,
swap_total,
cpu_temp,
sys_clock,
sys_clock,
mem_free,
mem_free,
mem_shared,
mem_shared,
mem_buffers,
mem_buffers,
cpu_idle,
cpu_idle,
swap_free,
swap_free,
load_one,
load_one,
load_five,
load_five,
load_fifteen,
load_fifteen,
proc_run,
proc_run,
proc_total,
proc_total, …..}
5
Metric.h
extern g_val_t cpu_num_func(void);
cpu_num_func(void);
extern g_val_t cpu_speed_func(void);
cpu_speed_func(void);
extern g_val_t mem_total_func(void);
mem_total_func(void);
extern g_val_t swap_total_func(void);
swap_total_func(void);
extern g_val_t sys_clock_func(void);
sys_clock_func(void);
extern g_val_t cpu_idle_func(void);
cpu_idle_func(void);
extern g_val_t load_one_func(void);
load_one_func(void);
extern g_val_t load_five_func(void);
load_five_func(void);
extern g_val_t load_fifteen_func(void);
load_fifteen_func(void);
extern g_val_t proc_run_func(void);
proc_run_func(void);
extern g_val_t proc_total_func(void);
proc_total_func(void);
extern g_val_t cpu_temp_func(void);
cpu_temp_func(void);
/machines/linux.c
g_val_t cpu_num_func ( void )
{
static int cpu_num = 0;
g_val_t val;
val; /* Only need to do this once */ if (! cpu_num)
cpu_num) {
cpu_num = get_nprocs();
get_nprocs();
}
val.uint16 = cpu_num;
cpu_num;
return val;
val;
}
g_val_t cpu_temp_func(void)
cpu_temp_func(void)
{
val.uint16=34;
return val;
val;
}
6
Using gmetric tool
Gmetric tool provides an easy way to
add metrics.
It is provided with Ganglia.
The metrics added by this tool do not
remain after a restart.
Syntax:
gmetric –-name=<metric name> --value=<valueofmetric>
--value=<valueofmetric>
--type=<typeofval>
--type=<typeofval> …
Example:
gmetric –-name=cpu_temp
name=cpu_temp –-value=30 –type=uint8
UC Berkeley Millennium Demo
Courtesy :http://monitor.millennium.berkeley.edu/
7
UC Berkeley Millennium Demo
Courtesy :http://monitor.millennium.berkeley.edu/
UC Berkeley Millennium Demo
8
RRDtool
RRDtool (Round Robin Database tool) is a
system to store and display time-series data
Creating RRD database
rrdtool create target.rrd --start
--start 1023654125 --step
--step 300
DS:mem:GAUGE:600:0:671744
Tutorial can be found at
http://people.ee.ethz.ch/~oetiker/webtools/rrdtool/tutorial/
RRDtool manuals at
http://people.ee.ethz.ch/~oetiker/webtools/rrdtool/manual/index.html
http://people.ee.ethz.ch/~oetiker/webtools/rrdtool/manual/index.html
