2
.\"___INFO__MARK_BEGIN__
4
.\" Copyright: 2004 by Sun Microsystems, Inc.
8
.\" $RCSfile$ Last Update: $Date$ Revision: $Revision$
11
.\" Some handy macro definitions [from Tom Christensen's man(1) manual page].
13
.de SB \" small and bold
14
.if !"\\$1"" \\s-2\\fB\&\\$1\\s0\\fR\\$2 \\$3 \\$4 \\$5
17
.de T \" switch to typewriter font
18
.ft CW \" probably want CW if you don't have TA font
21
.de TY \" put $1 in typewriter font
30
.de M \" man page reference
31
\\fI\\$1\\$2\\fR\\|(\\$3)\\$4
33
.TH xxQS_NAMExx_CKPT 1 "$Date$" "xxRELxx" "xxQS_NAMExx User Commands"
36
xxQS_NAMExx Checkpointing \- the xxQS_NAMExx checkpointing mechanism and checkpointing
41
supports two levels of checkpointing: the user level and a operating
42
system provided transparent
43
level. User level checkpointing refers to applications, which do their
44
own checkpointing by writing restart files at certain times or
45
algorithmic steps and by properly processing these restart files when
48
Transparent checkpointing has to be provided by the operating system and is
49
usually integrated in the operating system kernel. An example for a kernel
50
integrated checkpointing facility is the Hibernator package from Softway
51
for SGI IRIX platforms.
53
Checkpointing jobs need to be identified to the xxQS_NAMExx system by using the
54
\fI\-ckpt\fP option of the
56
command. The argument to this flag refers to a so
57
called checkpointing environment, which defines the attributes of the
58
checkpointing method to be used (see
61
Checkpointing environments are setup by the
63
options \fI\-ackpt\fP, \fI\-dckpt\fP, \fI\-mckpt\fP and \fI\-sckpt\fP. The
65
option \fI\-c\fP can be used to overwrite the \fIwhen\fP
66
attribute for the referenced checkpointing environment.
68
If a queue is of the type CHECKPOINTING, jobs need to have the
69
checkpointing attribute flagged (see the \fB\-ckpt\fP option to
71
to be permitted to run in such a queue. As opposed to the behavior for
72
regular batch jobs, checkpointing jobs are aborted under conditions,
73
for which batch or interactive jobs are suspended or even stay
74
unaffected. These conditions are:
77
Explicit suspension of the queue or job via
79
by the cluster administration or a queue owner
80
if the \fIx\fP occasion specifier (see
84
was assigned to the job.
87
A load average value exceeding the migration threshold as configured for
88
the corresponding queues (see
92
Shutdown of the xxQS_NAMExx execution daemon
93
.M xxqs_name_sxx_execd 8
94
being responsible for the checkpointing job.
97
After abortion, the jobs will migrate to other queues unless they were
98
submitted to one specific queue by an explicit user request.
99
The migration of jobs leads to a dynamic load balancing.
100
\fBNote:\fP The abortion of checkpointed jobs will free all resources
101
(memory, swap space) which the job occupies at that time. This is
102
opposed to the situation for suspended regular jobs, which still cover
108
When a job migrates to a queue on another machine at present no files
109
are transferred automatically to that machine. This means that all files
110
which are used throughout the entire job including restart files,
111
executables and scratch files must be visible or transferred explicitly
112
(e.g. at the beginning of the job script).
115
There are also some practical limitations regarding use of disk space
116
for transparently checkpointing jobs. Checkpoints of a transparently
117
checkpointed application are usually stored in a checkpoint file or
118
directory by the operating system. The file or directory contains all
119
the text, data, and stack space for the process, along with some
120
additional control information. This means jobs which use a very large
121
virtual address space will generate very large checkpoint files. Also
122
the workstations on which the jobs will actually execute may have
123
little free disk space. Thus it is not always possible to transfer a
124
transparent checkpointing job to a machine, even though that machine is
125
idle. Since large virtual memory jobs must wait for a machine that is
126
both idle, and has a sufficient amount of free disk space, such jobs
127
may suffer long turnaround times.
130
.M xxqs_name_sxx_intro 1 ,
135
.I xxQS_NAMExx Installation and Administration Guide,
136
.I xxQS_NAMExx User's Guide
140
.M xxqs_name_sxx_intro 1
141
for a full statement of rights and permissions.