9.2 Accounting: Job and System Statistics
Moab provides extensive accounting facilities that allow resource usage to be tracked by resources (compute nodes), jobs, users, and other objects. The accounting facilities may be used in conjunction with, and correlated with, the accounting records provided by the resource and allocation manager.
Moab maintains both raw persistent data and a large number of processed in memory statistics allowing instant summaries of cycle delivery and system utilization. With this information, Moab can assist in accomplishing any of the following tasks:
- Determining cumulative cluster performance over a fixed time frame.
- Graphing changes in cluster utilization and responsiveness over time.
- Identifying which compute resources are most heavily used.
- Charting resource usage distribution among users, groups, projects, and classes.
- Determining allocated resources, responsiveness, and failure conditions for jobs completed in the past.
- Providing real-time statistics updates to external accounting systems.
This section describes how to accomplish each of these tasks using Moab tools and accounting information.
9.2.1 Accounting Overview
Moab provides accounting data correlated to most major objects used within the cluster scheduling environment. These records provide job and reservation accounting, resource accounting, and credential based accounting.
|
9.2.1.1 Job and Reservation Accounting
As each job or reservation completes, Moab creates a complete persistent trace record containing information about who ran, the time frame of all significant events, and what resources were allocated. In addition, actual execution environment, failure reports, requested service levels, and other pieces of key information are also recorded. A complete description of each accounting data field can be found within section 16.3.3 Workload Traces.
9.2.1.2 Resource Accounting
The load on any given node is available historically allowing identification of not only its usage at any point in time, but the actual jobs which were running on it. Moab Cluster Manager can show load information (assuming load is configured as a generic metric), but not the individual jobs that were running on a node at some point in the past. For aggregated, historical statistics covering node usage and availability, the showstats command may be run with the -n flag.
9.2.1.3 Credential Accounting
Current and historical usage for users, groups, account, QoS's, and classes are determined in a manner similar to that available for evaluating nodes. For aggregated, historical statistics covering credential usage and availability, the showstats command may be run with the corresponding credential flag.
If needed, detailed credential accounting can also be enabled globally or on a credential by credential basis. With detailed credential accounting enabled, real-time information regarding per-credential usage over time can be displayed. To enable detailed per credential accounting, the ENABLEPROFILING attribute must be specified for credentials that are to be monitored. For example, to track detailed credentials, the following should be used:
USERCFG[DEFAULT] ENABLEPROFILING=TRUE QOSCFG[DEFAULT] ENABLEPROFILING=TRUE CLASSCFG[DEFAULT] ENABLEPROFILING=TRUE GROUPCFG[DEFAULT] ENABLEPROFILING=TRUE ACCOUNTCFG[DEFAULT] ENABLEPROFILING=TRUE
Credential level profiling operates by maintaining a number of time-based statistical records for each credential. The parameters PROFILECOUNT and PROFILEDURATION control the number and duration of the statistical records.
9.2.2 Real-Time Statistics
Moab provides real-time statistical information about how the machine is running from a scheduling point of view. The showstats command is actually a suite of commands providing detailed information on an overall scheduling basis as well as a per user, group, account and node basis. This command gets its information from in memory statistics that are loaded at scheduler start time from the scheduler checkpoint file. (See Checkpoint/Restart for more information.) This checkpoint file is updated periodically and when the scheduler is shut down allowing statistics to be collected over an extended time frame. At any time, real-time statistics can be reset using the resetstats command.
In addition to the showstats command, the showstats -f command also obtains its information from the in memory statistics and checkpoint file. This command displays a processor-time based matrix of scheduling performance for a wide variety of metrics. Information such as backfill effectiveness or average job queue time can be determined on a job size/duration basis.
9.2.3 FairShare Usage Statistics
Regardless of whether fairshare is enabled, detailed credential based fairshare statistics are maintained. Like job traces, these statistics are stored in the directory pointed to by the STATDIR parameter. Fairshare stats are maintained in a separate statistics file using the format FS.<EPOCHTIME> (FS.982713600, for example) with one file created per fairshare window. (See the Fairshare Overview for more information.) These files are also flat text and record credential based usage statistics. Information from these files can be seen via the mdiag -f command.
See Also