diff --git a/doc/html/power_mgmt.shtml b/doc/html/power_mgmt.shtml
index a3a2ed7600e93bbc6c3d10ae5b7ea14d9bcbe546..638ff857b5901e34e50634328bf83b7930162bc7 100644
--- a/doc/html/power_mgmt.shtml
+++ b/doc/html/power_mgmt.shtml
@@ -2,8 +2,18 @@
<h1>Slurm Power Management Guide</h1>
-<p>Slurm provied an integrated power management system for power capping.
-This is distict from Slurm's ability to
+<p>Slurm provide an integrated power management system for power capping.
+The mode of operation is to take the configured power cap for the system and
+distribute it across the compute nodes under Slurm control.
+Initially that power is distributed evenly across all compute nodes.
+Slurm then monitors actual power consumption and redistributes power as appropriate.
+Specifically, Slurm lowers the power caps on nodes using less then their cap
+and redistributes that power across the other nodes.
+The thresholds at which a node's power cap are raised or lowered are configurable
+as are the rate of change the power cap.
+In addition, starting a job on a node immediately triggers resetting the node's
+power cap to a higher level.
+Note this functionality is distinct from Slurm's ability to
<a href="power_save.html">power down idle nodes</a>.</p>
<h2>Configuration</h2>
@@ -13,24 +23,147 @@ This is distict from Slurm's ability to
<li><b>PowerParameters</b>:
Defines power management behavior.
+Changes to this value take effect when the Slurm daemons are reconfigured.
Currently valid options are:
<ul>
-<li>ClusterCap=#[KW|MW] - Used to establish cluster-wide power cap. Default
- units are watts.</li>
+<li><b>balance_interval=#</b> -
+ Specifies the time interval, in seconds, between attempts to balance power
+ caps across the nodes.
+ The default value is 30 seconds.
+ Supported by the power/cray plugin.</li>
+<li><b>capmc_path=</b> -
+ Specifies the absolute path of the <b>capmc</b> command.
+ The default value is "/opt/cray/capmc/default/bin/capmc".
+ Supported by the power/cray plugin.</li>
+<li><b>cap_watts=#[KW|MW]</b> -
+ Specifies the power limit to be established across all compute nodes managed
+ by Slurm.
+ The default value is no limit.
+ Supported by the power/cray plugin.</li>
+<li><b>decrease_rate=#</b> -
+ Specifies the maximum rate of change in the power cap for a node where the
+ actual power usage is below the power cap by an amount greater than
+ lower_threshold (see below).
+ Value represents a percentage of the difference between a node's minimum and
+ maximum power consumption.
+ The default value is 50 percent.
+ Supported by the power/cray plugin.</li>
+<li><b>increase_rate=#</b> -
+ Specifies the maximum rate of change in the power cap for a node where the
+ actual power usage is within upper_threshold (see below) of the power cap.
+ Value represents a percentage of the difference between a node's minimum and
+ maximum power consumption.
+ The default value is 20 percent.
+ Supported by the power/cray plugin.</li>
+<li><b>lower_threshold=#</b> -
+ Specify a lower power consumption threshold.
+ If a node's current power consumption is below this percentage of its current
+ cap, then its power cap will be reduced.
+ The default value is 90 percent.
+ Supported by the power/cray plugin.
+<li><b>recent_job=#</b> -
+ If a job has started or resumed execution (from suspend) on a compute node
+ within this number of seconds from the current time, the node's power cap will
+ be increased to the maximum.
+ The default value is 300 seconds.
+ Supported by the power/cray plugin.
+<li><b>upper_threshold=#</b> -
+ Specify an upper power consumption threshold.
+ If a node's current power consumption is above this percentage of its current
+ cap, then its power cap will be increased to the extent possible.
+ A node's power cap will also be increased if a job is newly started on it.
+ The default value is 95 percent.
+ Supported by the power/cray plugin.
</ul></li>
<li><b>PowerPlugin</b>:
Identifies the plugin used to manage system power consumption.
+Changes to this value require restarting Slurm daemons to take effect.
Currently valid options are:
<ul>
-<li>cray - Used for Cray systems with power monitoring and management
+<li><b>power/cray</b> -
+ Used for Cray systems with power monitoring and management
functionality included as part of System Management Workstation (SMW)
7.0.UP03.</li>
-<li>none - No power management support. The default value.</li>
+<li><b>power/none</b> - No power management support. The default value.</li>
</ul></li>
+</ul>
+
+<h2>Example</h2>
+
+<h3>Initial State</h3>
+<p>In our example, assume the following configuration:
+10 compute node cluster, where each node has a minimum power consumption of 100 watts
+and maximum power consumption of 200 watts.
+The following values for PowerParameters:
+balance_interval=60,
+cap_watts=1800,
+decrease_rate=30, increase_rate=10,
+lower_threshold=90, upper_threshold=98.
+The initial state is simply based upon the cap_watts divided by the number of
+compute nodes: 1800 watts / 10 nodes = 180 watts per node.</p>
+
+<h3>State in 60 Seconds</h3>
+<p>The power consumption is then examined balance_interval (60) seconds later.
+Assume that one of those nodes is consuming 110 watts and the others are
+using 180 watts.
+First we identify which nodes are consuming less than their lower_threshold
+of the power cap: 90% x 180 watts = 162 watts.
+One node falls in this category with 110 watts of power consumption.
+Its power cap is reduced by either half of the difference between it's current
+power cap and power consumption ((180 watts - 110 watts) / 2 = 35 watts) OR
+decrease_rate, which is a percentage of the difference between the node's
+maximum and minimum power consumption ((200 watts - 100 watts) x 30% = 30 watts).
+So that node's power cap is reduce from 180 watts to 150 watts.
+Ignoring the upper_threshold parameter for now, we now have 1650 watts available
+to distribute to the remaining 9 compute nodes, or 183 watts per node
+(1650 watts / 9 nodes = 183 watts per node).</p>
+<h3>State in 120 Seconds</h3>
+<p>The power consumption is then examined balance_interval (60) seconds later.
+Assume that one of those nodes is still consuming 110 watts, a second node is
+consuming 115 watts and the other eight are using 183 watts.
+First we identify which nodes are consuming less than their lower_threshold.
+Our node using 110 watts has its cap reduced by half of the difference between
+it's current power cap and power consumption
+((150 watts - 110 watts) / 2 = 20 watts);
+so that node's power cap is reduce from 150 watts to 130 watts.
+The node consuming 115 watts has its power cap reduced by 30 watts based
+decrease_rate; so that node's power cap is reduce from 183 watts to 153 watts.
+That leaves 1517 watts (1800 watts - 130 watts - 153 watts = 1517 watts) to
+be distributed over 8 nodes or 189 watts per node.</p>
+
+<h3>State in 180 Seconds</h3>
+<p>The power consumption is then examined balance_interval (60) seconds later.
+Assume the node previously consuming 110 watts is now consuming 128 watts.
+Since that is over upper_threshold of it's power cap
+(98% x 130 watts = 127 watts), its power cap is increased by increase_rate
+((200 watts - 100 watts) x 10% = 10 watts), so its power cap goes from
+130 watts to 140 watts.
+Assume the node previously consuming 115 watts has been allocated a new job.
+This triggers the node to be allocated the same power cap as nodes previously
+running at their power cap.
+Therefore we have 1660 watts available (1800 watts - 140 watts = 1660 watts)
+to be distributed over 9 nodes or 184 watts per node.</p>
+
+<h2>Notes</h2>
+<ul>
+<li>We need to determine how the current power consumption data is gathered
+ and determine accuracy of that data.</li>
+<li>The current algorithm manages power on a per-node basis without considering
+ job allocations. Should an attempt be made to maintain uniform power caps
+ across all nodes associated with each job? Doing so could greatly complicate
+ the logic.</li>
+<li>Current default values for configuration parameters should probably be
+ changed once we have a better understanding of the algorithm's behavior.</li>
+<li>No integration of this logic with gang scheduling currently exists.
+ It is not clear that configuration is practical to support as gang scheduling
+ time slices will typically be smaller than the power management
+ balance_interval.</li>
+<li>No integration with logic powering down idle nodes is currently implemented.
+ That should be practical to add, but has not yet been studied.</li>
</ul>
-<p style="text-align:center;">Last modified 20 January 2015</p>
+<p style="text-align:center;">Last modified 28 January 2015</p>
<!--#include virtual="footer.txt"-->
diff --git a/doc/man/man5/slurm.conf.5 b/doc/man/man5/slurm.conf.5
index 5bf3f42f35b88901a40dfbec79f234c9bffe3995..a366bdce95ad5ff90e915eb0860996becf2f2278 100644
--- a/doc/man/man5/slurm.conf.5
+++ b/doc/man/man5/slurm.conf.5
@@ -1402,6 +1402,7 @@ on SPANK plugins, see the \fBspank\fR(8) manual.
\fBPowerParameters\fR
System power management parameters.
The supported parameters are specific to the \fBPowerPlugin\fR.
+Changes to this value take effect when the Slurm daemons are reconfigured.
More information about system power management is available here
<http://slurm.schedmd.com/power_mgmt.html>.
Options current supported by any plugins are listed below.
@@ -1438,7 +1439,7 @@ Specifies the maximum rate of change in the power cap for a node where the
actual power usage is within \fBupper_threshold\fR (see below) of the power cap.
Value represents a percentage of the difference between a node's minimum and
maximum power consumption.
-The default value is 10 percent.
+The default value is 20 percent.
Supported by the power/cray plugin.
.TP
\fBlower_threshold=#\fR
@@ -1448,6 +1449,13 @@ cap, then its power cap will be reduced.
The default value is 90 percent.
Supported by the power/cray plugin.
.TP
+\fBrecent_job=#\fR
+If a job has started or resumed execution (from suspend) on a compute node
+within this number of seconds from the current time, the node's power cap will
+be increased to the maximum.
+The default value is 300 seconds.
+Supported by the power/cray plugin.
+.TP
\fBupper_threshold=#\fR
Specify an upper power consumption threshold.
If a node's current power consumption is above this percentage of its current
@@ -1462,6 +1470,7 @@ Identifies the plugin used for system power management.
Currently supported plugins include:
\fBcray\fR and
\fBnone\fR (default).
+Changes to this value require restarting Slurm daemons to take effect.
More information about system power management is available here
<http://slurm.schedmd.com/power_mgmt.html>.
diff --git a/src/common/power.c b/src/common/power.c
index 8a2b8a3a7d7869e3636de853e4003872ed642908..63ffd2f9de07a49602fafe0da26af1f81103ed2d 100644
--- a/src/common/power.c
+++ b/src/common/power.c
@@ -54,7 +54,9 @@
* working. If you need to add fields, add them at the end of the structure.
*/
typedef struct slurm_power_ops {
+ void (*job_resume) (struct job_record *job_ptr);
void (*job_start) (struct job_record *job_ptr);
+ void (*reconfig) (void);
} slurm_power_ops_t;
/*
@@ -62,7 +64,9 @@ typedef struct slurm_power_ops {
* declared for slurm_power_ops_t.
*/
static const char *syms[] = {
- "power_p_job_start"
+ "power_p_job_resume",
+ "power_p_job_start",
+ "power_p_reconfig"
};
static int g_context_cnt = -1;
@@ -72,6 +76,7 @@ static char *power_plugin_list = NULL;
static pthread_mutex_t g_context_lock = PTHREAD_MUTEX_INITIALIZER;
static bool init_run = false;
+/* Initialize the power plugin */
extern int power_g_init(void)
{
int rc = SLURM_SUCCESS;
@@ -125,6 +130,7 @@ fini:
return rc;
}
+/* Terminate the power plugin and free all memory */
extern int power_g_fini(void)
{
int i, j, rc = SLURM_SUCCESS;
@@ -151,6 +157,30 @@ fini: slurm_mutex_unlock(&g_context_lock);
return rc;
}
+/* Read the configuration file */
+extern void power_g_reconfig(void)
+{
+ int i;
+
+ (void) power_g_init();
+ slurm_mutex_lock(&g_context_lock);
+ for (i = 0; i < g_context_cnt; i++)
+ (*(ops[i].reconfig))();
+ slurm_mutex_unlock(&g_context_lock);
+}
+
+/* Note that a suspended job has been resumed */
+extern void power_g_job_resume(struct job_record *job_ptr)
+{
+ int i;
+
+ (void) power_g_init();
+ slurm_mutex_lock(&g_context_lock);
+ for (i = 0; i < g_context_cnt; i++)
+ (*(ops[i].job_resume))(job_ptr);
+ slurm_mutex_unlock(&g_context_lock);
+}
+
/* Note that a job has been allocated resources and is ready to start */
extern void power_g_job_start(struct job_record *job_ptr)
{
diff --git a/src/common/power.h b/src/common/power.h
index eb34e5d1bcf5ca6f035d0a9f1749e7037f6c99e5..6eaa731608e6154619d9a871a38d4273d3b9cb15 100644
--- a/src/common/power.h
+++ b/src/common/power.h
@@ -48,11 +48,17 @@
/*****************************************************************************\
* PLUGIN FUNCTIONS
\*****************************************************************************/
-/* initialize the power plugin */
-extern int power_g_init (void);
+/* Initialize the power plugin */
+extern int power_g_init(void);
-/* terminate the power plugin and free all memory */
-extern int power_g_fini (void);
+/* Terminate the power plugin and free all memory */
+extern int power_g_fini(void);
+
+/* Read the configuration file */
+extern void power_g_reconfig(void);
+
+/* Note that a suspended job has been resumed */
+extern void power_g_job_resume(struct job_record *job_ptr);
/* Note that a job has been allocated resources and is ready to start */
extern void power_g_job_start(struct job_record *job_ptr);
diff --git a/src/plugins/power/cray/power_cray.c b/src/plugins/power/cray/power_cray.c
index ad3b0aa3d548b14ba15f444ed96e2ca955f81c6d..f71dfdc67bdf6e5b92fceb1bd23445594fefe8de 100644
--- a/src/plugins/power/cray/power_cray.c
+++ b/src/plugins/power/cray/power_cray.c
@@ -63,7 +63,7 @@
#define DEFAULT_CAPMC_PATH "/opt/cray/capmc/default/bin/capmc"
#define DEFAULT_CAP_WATTS 0
#define DEFAULT_DECREASE_RATE 50
-#define DEFAULT_INCREASE_RATE 10
+#define DEFAULT_INCREASE_RATE 20
#define DEFAULT_LOWER_THRESHOLD 90
#define DEFAULT_UPPER_THRESHOLD 95
#define DEFAULT_RECENT_JOB 300
@@ -128,6 +128,7 @@ static uint64_t debug_flag = 0;
static uint32_t decrease_rate = DEFAULT_DECREASE_RATE;
static uint32_t increase_rate = DEFAULT_INCREASE_RATE;
static uint32_t lower_threshold = DEFAULT_LOWER_THRESHOLD;
+static uint32_t recent_job = DEFAULT_RECENT_JOB;
static uint32_t upper_threshold = DEFAULT_UPPER_THRESHOLD;
static bool stop_power = false;
static pthread_t power_thread = 0;
@@ -217,6 +218,15 @@ static void _load_config(void)
}
}
+ if ((tmp_ptr = strstr(sched_params, "recent_job="))) {
+ recent_job = atoi(tmp_ptr + 11);
+ if (recent_job < 1) {
+ error("PowerParameters: recent_job=%u invalid",
+ recent_job);
+ recent_job = DEFAULT_RECENT_JOB;
+ }
+ }
+
if ((tmp_ptr = strstr(sched_params, "upper_threshold="))) {
upper_threshold = atoi(tmp_ptr + 16);
if (upper_threshold < 1) {
@@ -230,9 +240,10 @@ static void _load_config(void)
if (debug_flag & DEBUG_FLAG_POWER) {
info("PowerParameters=balance_interval=%d,capmc_path=%s,"
"cap_watts=%u,decrease_rate=%u,increase_rate=%u,"
- "lower_threashold=%u,upper_threshold=%u",
+ "lower_threashold=%u,recent_job=%u,upper_threshold=%u",
balance_interval, capmc_path, cap_watts, decrease_rate,
- increase_rate,lower_threshold, upper_threshold);
+ increase_rate, lower_threshold, recent_job,
+ upper_threshold);
}
}
@@ -352,6 +363,24 @@ static void _json_parse_object(json_object *jobj, power_config_nodes_t *ent)
}
}
+static void _my_sleep(int add_secs)
+{
+ struct timespec ts = {0, 0};
+ struct timeval tv = {0, 0};
+
+ if (gettimeofday(&tv, NULL)) { /* Some error */
+ sleep(1);
+ return;
+ }
+
+ ts.tv_sec = tv.tv_sec + add_secs;
+ ts.tv_nsec = tv.tv_usec * 1000;
+ pthread_mutex_lock(&term_lock);
+ if (!stop_power)
+ pthread_cond_timedwait(&term_cond, &term_lock, &ts);
+ pthread_mutex_unlock(&term_lock);
+}
+
/* Periodically attempt to re-balance power caps across nodes */
extern void *_power_agent(void *args)
{
@@ -367,7 +396,7 @@ extern void *_power_agent(void *args)
last_balance_time = time(NULL);
while (!stop_power) {
- sleep(1);
+ _my_sleep(1);
if (stop_power)
break;
@@ -412,7 +441,7 @@ static List _rebalance_node_power(void)
struct node_record *node_ptr, *node_ptr2;
uint32_t alloc_power = 0, avail_power, ave_power, new_cap, tmp_u32;
int node_power_raise_cnt = 0;
- time_t recent = time(NULL) - DEFAULT_RECENT_JOB;
+ time_t recent = time(NULL) - recent_job;
int i, j;
/* Lower caps on under used nodes */
@@ -656,6 +685,22 @@ extern void fini(void)
pthread_mutex_unlock(&thread_flag_mutex);
}
+/* Read the configuration file */
+extern void power_p_reconfig(void)
+{
+ pthread_mutex_lock(&thread_flag_mutex);
+ _load_config();
+ if (cap_watts == 0)
+ _stop_power_agent();
+ pthread_mutex_unlock(&thread_flag_mutex);
+}
+
+/* Note that a suspended job has been resumed */
+extern void power_p_job_resume(struct job_record *job_ptr)
+{
+ set_node_new_job(job_ptr, node_record_table_ptr);
+}
+
/* Note that a job has been allocated resources and is ready to start */
extern void power_p_job_start(struct job_record *job_ptr)
{
diff --git a/src/plugins/power/none/power_none.c b/src/plugins/power/none/power_none.c
index f39a2653c80624de56b07939b9f2689e2ffbb994..933bc927780aff74a6e84aab66ff98894938ca14 100644
--- a/src/plugins/power/none/power_none.c
+++ b/src/plugins/power/none/power_none.c
@@ -99,6 +99,18 @@ extern int fini(void)
return SLURM_SUCCESS;
}
+/* Read the configuration file */
+extern void power_p_reconfig(void)
+{
+ return;
+}
+
+/* Note that a suspended job has been resumed */
+extern void power_p_job_resume(struct job_record *job_ptr)
+{
+ return;
+}
+
/* Note that a job has been allocated resources and is ready to start */
extern void power_p_job_start(struct job_record *job_ptr)
{
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index a14002fcb53d67b682cbb34411386ebbf3987a0f..311aff7c2d275c5391cfe82a02238e407abf8507 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -71,6 +71,7 @@
#include "src/common/hostlist.h"
#include "src/common/node_select.h"
#include "src/common/parse_time.h"
+#include "src/common/power.h"
#include "src/common/slurm_accounting_storage.h"
#include "src/common/slurm_jobcomp.h"
#include "src/common/slurm_priority.h"
@@ -12434,6 +12435,7 @@ static int _job_suspend(struct job_record *job_ptr, uint16_t op, bool indf_susp)
if (!IS_JOB_SUSPENDED(job_ptr))
return ESLURM_JOB_NOT_SUSPENDED;
rc = _resume_job_nodes(job_ptr, indf_susp);
+ power_g_job_resume(job_ptr);
if (rc != SLURM_SUCCESS)
return rc;
_suspend_job(job_ptr, op, indf_susp);
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index 74d86eded275006986aeaf4517da732741cc1a87..911d0f8031eb2e6cb3bb859397d2fbd99fe7106e 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -67,14 +67,15 @@
#include "src/common/macros.h"
#include "src/common/node_select.h"
#include "src/common/parse_spec.h"
+#include "src/common/power.h"
#include "src/common/read_config.h"
#include "src/common/slurm_jobcomp.h"
#include "src/common/slurm_topology.h"
#include "src/common/slurm_rlimits_info.h"
#include "src/common/slurm_route.h"
+#include "src/common/strnatcmp.h"
#include "src/common/switch.h"
#include "src/common/xstring.h"
-#include "src/common/strnatcmp.h"
#include "src/slurmctld/acct_policy.h"
#include "src/slurmctld/burst_buffer.h"
@@ -1006,6 +1007,7 @@ int read_slurm_conf(int recover, bool reconfig)
rehash_node();
slurm_topo_build_config();
route_g_reconfigure();
+ power_g_reconfig();
cpu_freq_reconfig();
rehash_jobs();