load_parameters - Grid Engine default load parameters
The load parameters reported by the execution daemon
sge_execd(8) by default currently are:
- arch
- An architecture string compiled into the sge_execd describing the
operating system architecture for which the execd is targeted. Possible
values include:
aix51 for IBM AIX 5.1, 5.3
darwin-x86 for Apple Mac OS/X x86
darwin-ppc for Apple Mac OS/X PowerPC
hp11 for HP HP-UX 11.x 32-bit
hp11-64 for HP HP-UX 11.x 64-bit
lx-amd64 for GNU/Linux x86_64
lx-x86 for GNU/Linux x86
sol-sparc64 for Sun Solaris Sparc64
sol-x86 for Sun Solaris x86
sol-amd64 for Sun Solaris x86_64
win32-x86 for Windows x86
Note that an sge_execd for a particular architecture may run on multiple OS
versions. In this case, the architecture string delivered by the execution
daemon may be misleading.
- num_proc
- The number of processors provided by the execution host. The host is
defined by a single Internet address, i.e. rack mounted multi host systems
are counted as a cluster rather than a single multi headed machine. In the
case of "hardware threads" ("CMT" or "SMT"),
normally num_proc is the number of such threads which are activated.
The following load information is only available under IBM AIX if
Grid Engine is built with the perfstat library (which is now the
default):
- load_short
- The short time average OS run queue length. It is the first of the value
triple reported by uptime(1). Many implementations provide a 1
minute average with this value.
- load_medium
- The medium time average OS run queue length. It is the second of the value
triple reported by uptime(1). Many implementations provide a 5
minute average with this value.
- load_long
- The long time average OS run queue length. It is the third of the value
triple reported by uptime(1). Many implementations provide a 10 or
15 minute average with this value.
- load_avg
- The same as load_medium.
- np_load_short
- The same as load_short but divided by the number of processors. This value
allows comparing the load of single and multi-headed hosts.
- np_load_medium
- The same as load_medium but divided by the number of processors. This
value allows comparing the load of single and multi-headed hosts.
- np_load_long
- The same as load_long but divided by the number of processors. This value
allows comparing the load of single and multi-headed hosts.
- np_load_avg
- The same as load_avg but divided by the number of processors. This value
allows comparing the load of single and multi-headed hosts.
- cpu
- The percentage of CPU time not in idle state.
The following load information is only available under IBM AIX if
Grid Engine is built with the perfstat library (which is now the
default):
- mem_free
- The amount of free real memory.
- swap_free
- The amount of free swap memory.
- virtual_free
- The sum of mem_free and swap_free.
- mem_used
- The amount of memory used.
- swap_used
- The amount of swap space used.
- virtual_used
- The sum of mem_used and swap_used.
- mem_total
- The total amount of memory (free+used).
- swap_total
- The total amount of swap space (free+used).
- virtual_total
- The sum of mem_total and swap_total.
The following load information is ONLY available under
Windows:
- display_win_gui
- The host is capable of displaying the GUI of a Windows job.
The following load information is only available if Grid Engine
was built with the hwloc library (the default) and if hwloc supports the
system.
- m_topology_inuse
- The host cpu topology string reported by an execution host. This might be
"NONE" if the topology cannot be determined; otherwise it is a
string consisting of the upper and lowercase letters "S",
"C", "T", "c", and "t". The
sequence of letters within that string represents the hardware topology
where "S" represents a socket, "C" or "c" a
core, and "T" or "t" a hardware thread.
The string "SCCSCCSCCSCC" will returned by a host
that has 4 sockets where each of those sockets has two cores. All cores
are available because all "C"s appear in capital letters.
If lowercase letters are used then this means that the
corresponding core or thread is already in use because there is at least
one running Grid Engine job bound to it.
"SCCSCcSCCscc" means that core 2 on socket 2 and
also core 1 and core 2 on socket 4 are in use.
- m_topology
- Like m_topology_inuse, except that it represents the architecture, not the
use of it, i.e. it doesn't contain lower case letters.
- m_socket
- Number of sockets available on the reporting host (e.g.
"SCCSCCSCCSCC" => m_socket=4)
- m_core
- Number of cores reported for all sockets on a host (e.g.
"SCCSCCSCCSCC" => m_core=2)
- m_thread
- Number of hardware threads reported for all cores on a host, which should
be equal to num_proc (e.g. "SCTTCTTCTTCTT" =>
m_thread=8)
The "static" load values (which may be selected with
qhost -l or qselect -l) are: arch, num_proc, mem_total,
swap_total, virtual_total, m_topology, m_socket, m_core, m_thread.