US Patent No. 9,917,797

METHOD AND SYSTEM FOR MANAGING SWITCH WORKLOADS IN A CLUSTER


Patent No. 9,917,797
Issue Date March 13, 2018
Title Method And System For Managing Switch Workloads In A Cluster
Inventorship Benoit Sigoure, San Francisco, CA (US)
Assignee Arista Networks, Inc., Santa Clara, CA (US)

Claim of US Patent No. 9,917,797

1. A switch, comprising:
a processor;
a plurality of ports;
memory comprising instructions, which when executed by the processor perform a method, the method comprising:
obtaining, via a port of the plurality of ports, a global task-server mapping from a first server, wherein the global task-server
mapping specifies which of a plurality of tasks is currently executing on which of a plurality of servers in a first cluster;

detecting that one of the plurality of ports is congested;
based on the detecting:
identifying a task associated with the one of the plurality of ports using the global task-server mapping;
generating a task listing based on the identifying, wherein the task listing specifies at least the task executing on a server,
of the plurality of servers, directly connected to the one of the plurality of ports;

generating an alert comprising the task listing;
transmitting the alert to an administrator;
detecting that a link down has occurred on a second port of the plurality of ports, wherein the second port is directly connected
to a DataNode server comprising stored data;

sending, based on detecting the link down, a link down message to a NameNode server,
wherein the NameNode server, in response to receiving the link down message:
identifies a plurality of other DataNode servers capable of receiving copies the stored data,
identifies, using global switch load information, a subset of the plurality of other DataNode servers, and
replicates a copy of the stored data to each of the DataNode servers in the subset.