Browse Prior Art Database

System and Method to Bias Workload Distribution around Uplink Constraints

IP.com Disclosure Number: IPCOM000238591D
Publication Date: 2014-Sep-04
Document File: 2 page(s) / 102K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed are a system and method to distribute workloads within a network topology by optimizing the use of uplinks to minimize cross-switch conflicts and latency.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 01 of 2

System and Method to Bias Workload Distribution around Uplink Constraints

Distribution of jobs is constantly getting smarter in an effort to make the best use of computing resources. Various cloud and high performance computing (HPC) providers continually search for innovations to products that drive sales. One significant area is optimization around the network topology of an HPC/cloud/data center; particularly, where blocking ratios exist within the network or where Message Passing Interfaces (MPI) or other communications may result in saturation of the uplinks on any given switch or fabric.

Figure 1: Typical job distribution with bottleneck created by workload distribution that does not consider topology (48 nodes communicating over 24 uplinks)

A method is needed to optimize placement and distribution of workloads to optimize the usage of uplinks in a network topology.

The solution is a novel and non-obvious system and method to distribute workloads in a manner that optimizes use of uplinks to minimize cross-switch conflicts and latency. In implementation, this exists as software/service in a management entity or workload manager.

To operate the system and method in a preferred embodiment:

1. Assess topology of a network

A. Query switches (Simple Network Management Protocol (SNMP), etc.) for upward and downward bound links

    B. Measure bandwidth/packet loss across or between points/systems 2. Determine optimal workload placement
A. Determine or receive an estimated number of n...