Unterschiede

Hier werden die Unterschiede zwischen zwei Versionen angezeigt.

--- hpc:numericserveruser:hpc_numericserver:batch_job_facility:queues_numericserver [2019/06/19 19:32] – kraus
+++ hpc:numericserveruser:hpc_numericserver:batch_job_facility:queues_numericserver [2019/11/20 00:28] (aktuell) – [Defining parallel job environments (PE)] kraus
@@ Zeile 1: / Zeile 1: @@
+======= Job queues and parallel job environments at Numerikserver =======
+===== Preliminary notes about the architecture of the Numerikserver =====
+  * SGE knows about several special environments for parallel jobs types, the PE. Each process of a parallel job is bound to one slot. In the current setup, a slot corresponds to a processor core.
+  * <fc #ff0000>Parallel jobs need to specify a corresponding parallel environment within their bach job script.</fc>
+  * The Numerikserver divides into 21 clusters, each containing nodes with a specific hardware architecture.
+\\
+===== About queues and parallel environments in general =====
+  * Queues allow to classify jobs with respect to their run-time (wall clock time).
+  * Parallel Environments (PE) allow to classify jobs with respect their type of parallelism.
+\\
+===== Query list of available queues =====
+  * type the command: <code>qconf -sql</code>
+===== Query properties of a single queue =====
+  * type the command: <code>qconf -sq [name_of_queue]</code>
+\\
+===== Available queues - different run-time limits =====
+  * IMPORTANT NOTE: <fc #ff0000>Depending on the run-time requirements of your job, the appropriate queue will be automatically chosen. No need to specify the queue explicitly.</fc>
+  * **short**: 12 h
+  * **medium**:72 hours (3 d)
+  * **long**: 220 hours (9 d 4 h) \\
+  * **verylong** (only on three clusters): 1344 hours (56 d)
+\\
+\\
+===== Defining parallel job environments (PE) =====
+  * General syntax defining a PE: ''[mp|ompi]<clusterid|*>_<noprocs-per-node> [number_of_proceses|range_number_of_processes]'' \\ (mandatory arguments in ''[]'' and optional arguments in ''<>'' given)
+  * Several examples:
+    * SMP/MPI job intranode (request 9 slots on a single node within any cluster): ''mp 9''
+    * MPI job internode (request 18 slots on a variable number of nodes residing in any cluster; NB: all nodes reside within the same cluster) : ''ompi 18'' or ''ompi* 18''
+    * MPI job internode on a specific cluster (request 20 slots on a variable number of nodes residing in cluster 4): ''ompi4 20''
+    * MPI job internode with explicit partitioning (request in total 32 slots, on each node spending 8 slots): ''ompi*_8 32''
+\\
+===== Query list of available PE =====
+  * type the command: <code>qconf -spl</code>
+===== Query properties of a single PE =====
+  * type the command: <code>qconf -sp [name_of_PE]</code>
+\\
+===== Query resources of a specific cluster node =====
+  * type the command: <code>qhost</code>
+\\
+<WRAP rightalign>
+date of revision: 06-19-2019 © kraus
+</WRAP>