Skip to content

Queues

Different task scheduling Queues are available in order to facilitate flexible access to the computational resources.

Naming Convention

Queue name consists of the 4 main parts as demonstrated below. Only one part, denoted as 2 and enclosed by braces {}, is mandatory. Other parts, enclosed by brackets [], are optional.

1
[1a][1b][1c]{2}[3][4]

Explanation

  1. GPU-specific concerns:

    a. Whether queue is GPU-enabled. If yes, letter "G" is used. b. Type of GPU. P means P100. V100 is used by default for GPU-enabled queues otherwise. c. Number of GPUs per each node, 1 GPU if not specified.

  2. Queue cost category: either "D" (Debug), "O" (Ordinary) or "S" (Saving).

  3. Queue provision mode, R (Regular) and F (Fast).

  4. Maximum number of cores per single node. Depends on the cluster and cloud provider if it is not specified.

Examples

  • G4OF: GPU-enabled, V100, 4 GPUs, Ordinary, Fast; 32 cores per node on AWS only
  • GPSF, GPU-enabled, P100, 1 GPU, Saving, Fast, 6 cores per node on Azure only
  • OR16: Ordinary, Regular, 16 cores per each compute node
  • OF: Ordinary, Fast; 36 cores per node on AWS, 16 cores on Azure
  • SF+: Saving, Fast; 72 later generation (compared to "SF") cores on AWS only

Charge Policies

We deploy two charge policies, as explained below:

  • core seconds: jobs are charged according to the number of core-seconds consumed
  • core hours: jobs are charged according to the nearest (greater) integer number of core-hours consumed; also referred to as "whole hours"

The latter is used for queues with Fast provision category. The former is used otherwise.

Be considerate when using queues with core-hours charge policies

When tasks are submitted to the queue with "core hours" based charge policy, our accounting system would charge the account for at least 1 hour of usage of the resource. We advise users to prototype the calculations in other queues and deploy production-ready large-scale runs using queues with "core hours" charge policies.

List of Queues

Detailed cluster-specific lists of queues are available for the Amazon Web Services and Azure separately.

Select Queue

Queues can be selected under the Web Interface according to the instructions found here. Similarly, the desired queue can be specified in the Batch Script for the case of Job submission via the Command Line Interface.