array jobs are not included
array jobs are not included
MetaCentrum operating statistics include overviews of calculated time and the number of tasks processed on machines integrated into MetaCentrum, waiting times and task runtimes, institutional composition, applications used, most active users, machine utilization within MetaCentrum, and the usage of open-access clusters and machines by institutions.
Processors/CPUs: Only physical processor cores are counted. For example, a quad-core chip is counted as four CPUs. A single processor with two HyperThreading cores is counted as one processor. A dual-core chip where each core has two HyperThreading cores is counted as two processors.
Task Time: This refers to the task duration within the monitored period. For tasks started before the beginning of the period, only the portion from the start of the period is counted. Conversely, for tasks still running at the end of the monitored period, only the portion preceding the end of the period is included.
Application Statistics Statistics for the applications used were obtained from Linux kernel accounting records regarding processes. All processes from all MetaCentrum machines lasting longer than one second are recorded. For each process, the name of the executed file (without the file path) and other data—specifically utime, stime, and etime (user time, system time, elapsed time)—are known.
The names of executed files were compared against regular expressions capturing the filenames of known applications. For example, any process matching the following regular expression is categorized under the Amber application:
ptraj|pmemd.*|_pmemd.*|_sander.*|sander.*|teLeap|xaLeap|mm_pbsa.pl|pbsa|nmode|mopac|pmtree_.*|alltoall|analysis|normalization
The utime and stime of all processes for each application were then summed. Thus, the statistics represent the time actually consumed by calculations.
Machine Utilization Utilization is calculated by individual clusters because MetaCentrum consists of diverse machines belonging to different organizations. Since some machines and clusters are fully reserved for their owners, aggregates for the entirety of MetaCentrum would be misleading.
Utilization data is drawn from the PBS scheduling system task logs. Processors assigned to a specific task are considered fully utilized from the start to the end of that task. Therefore, utilization statistics show the time allocated to users, rather than how effectively the users utilize that allocated time themselves.
Utilization is calculated separately for each day:
Baseline: The number of CPUs in a given cluster at the start of the day serves as the baseline; changes in CPU counts are reflected the following day.
CPU-Seconds: The number of CPUs is multiplied by the seconds in that day (usually 86,400, except for Daylight Savings transitions). This total of CPU-seconds is the reference value for that cluster and day.
Every CPU-second is categorized into one of the following states:Allocated: Assigned to a task.
Maintenance: The machine was in a maintenance queue or marked as down in the PBS logs.
Reserved: Reserved by means other than a standard task (e.g., placed in a reserved queue, designated as a frontend, or marked as reserved in the Perun system—this applies to nodes like skirit, manwe3, manwe4, wood, oskarnode, etc.).
Unused: Idle capacity.