Node selection for message redistribution in an integrated application-aware load balancer incorporated within a distributed-service-application-controlled distributed computer system

Patent metadata

Field	Value
Publication number	US-10320891-B2
Application number	US-201615005922-A
Country	US
Kind code	B2
Filing date	Jan 25, 2016
Priority date	Jan 25, 2016
Publication date	Jun 11, 2019
Grant date	Jun 11, 2019

Abstract

The current document is directed to an integrated application-aware load-balancing component of a distributed computer system. The integrated application-aware load-balancing component is, in a described embodiment, incorporated within a distributed application that serves as a control component of multiple physical computers within a distributed computer system. The integrated application-aware load-balancing component includes a layer-4 load-balancing subcomponent that distributes communications connections initiated by remote client computers among computational nodes within the distributed computer system in order to balance the computational load applied to the distributed application and distributed computer system by the remote client computers. The integrated application-aware load-balancing component additionally includes a layer-7 load-balancing subcomponent that periodically and locally computes weights for each computational node and redistributes client requests according to the computed weights among cluster nodes within the distributed computer system in order to enhance balancing of the computational load applied to the distributed service application and distributed computer system by the remote client computers. Operation of the layer-4 and layer-7 load balancers are controlled and/or adjusted by rules and policies informed by performance and operational characteristics of the distributed service application and distributed computer system.

First claim

The invention claimed is:

1. An integrated, application-aware load-balancing component of a distributed computer system controlled by a distributed application, the integrated, application-aware load-balancing component comprising:
the distributed computer system having multiple computational nodes, each controlled by a local instance of the distributed application that includes a local instance of the integrated, application-aware load-balancing component;
a layer-4 load-balancing subcomponent that
accesses a connection table that stores entries that each represents a current connection between a client device and the multiple computational nodes,
accesses a set of stored layer-4 load-balancing policies, and
distributes, in accordance with the set of stored layer-4 load-balancing policies, communications connections from remote processor-controlled client devices to the distributed application among the multiple computational nodes in order to balance the computational load applied to the distributed computer system by the remote processor-controlled client devices, updating the connection table when a new communications connection is established; and

a layer-7 load-balancing subcomponent that
accesses a set of stored layer-7 load-balancing policies, and
periodically computes weights for each computational node and uses the computed weights to redistribute, in accordance with the set of stored layer-4 load-balancing policies, messages transmitted to the distributed computer system by the remote processor-controlled client devices, directed to the distributed application, and received by one or more of the multiple computational nodes among one or more of the multiple computational nodes in order to balance the computational load applied to the distributed computer system by the remote processor-controlled client devices;

wherein each of the multiple computational nodes includes a local layer-7 load-balancing subcomponent; and
wherein each local layer-7 load-balancing subcomponent computes weights for each computational node of the multiple computational nodes at points in time separated by one or more weight-computing time intervals based on a number of layer-4-traffic messages received by the computational node that includes the local layer-7 load-balancing subcomponent during an immediately preceding accumulation time interval that is longer than the weight-computing time intervals.

2. The integrated, application-aware load-balancing component of claim 1 wherein each local layer-7 load-balancing subcomponent computes weights for each computational node by:
assigning target weights to the multiple computational nodes;
determining a fair share for each computational node of the multiple computational nodes based on the number of layer-4-traffic messages received by the multiple computational nodes during an immediately preceding accumulation time interval and the number of the multiple computational nodes; and
when the local computational node that includes the local layer-7 load-balancing subcomponent received, during the immediately preceding accumulation time interval, more than the determined fair share of layer-4-traffic messages,
decreasing the weight assigned to the local computational node, and
increasing the weights assigned to those computational nodes, of the remaining computational nodes of the multiple computational nodes which do not include the local computational node, that received, during the immediately preceding accumulation time interval, fewer than the determined fair share of layer-4-traffic messages.

3. The integrated, application-aware load-balancing component of claim 2 wherein each weight assigned to a computational node by each local layer-7 load-balancing subcomponent is a floating-point value in the range from 0 to 1, denoted [0, 1], and wherein the weights assigned to the computational nodes are normalized so that a sum of the weights assigned to the computational nodes is either 1 or falls in a range selected from among:
[1, 1];
[0.99, 1.01];
[0.999, 1.001]; and
other ranges that include 1 and that span a continuous length of less than 0.1.

4. The integrated, application-aware load-balancing component of claim 2 wherein the local layer-7 load-balancing subcomponent decreases the weight assigned to the local computational node, when the local computational node that includes the local layer-7 load-balancing subcomponent received, during the immediately preceding accumulation time interval, more than the determined fair share of layer-4-traffic messages, to a ratio of the fair share divided by the number of messages received by the local computational node during the immediately preceding accumulation time interval.

5. The integrated, application-aware load-balancing component of claim 2 wherein the local layer-7 load-balancing subcomponent increases the weights assigned to each of those computational nodes of the remaining computational nodes that received, during the immediately preceding accumulation time interval, fewer than the determined fair share of layer-4-traffic messages, referred to as “underperforming node,” by:
computing, for each underperforming node, a shortage as the number of messages received by the underperforming node during the immediately preceding accumulation time interval subtracted from the fair share;
computing the sum of the computed shortages; and
for each underperforming node, dividing the shortage computed for the underperforming node by the sum of the computed shortages to produce a first intermediate value that is multiplied by a second intermediate value computed as the weight assigned to the local node subtracted from 1.

6. The integrated, application-aware load-balancing component of claim 2 wherein, when the local computational node did not receive, during the immediately preceding accumulation time interval, more than the determined fair share of layer-4-traffic messages, the local load-balancing subcomponent decreases the weights assigned to the remaining computational nodes and increases the weight assigned to the local computational node.

7. The integrated, application-aware load-balancing component of claim 6 wherein the local layer-7 load-balancing subcomponent decreases the weight assigned to each of the remaining computational nodes when the local computational node did not receive, during the immediately preceding accumulation time interval, more than the determined fair share of layer-4-traffic messages by setting the weight assigned to the remaining computational node to 0.

8. The integrated, application-aware load-balancing component of claim 2 wherein the local layer-7 load-balancing subcomponent increases the weight assigned to the local computational node when the local computational node did not receive, during the immediately preceding accumulation time interval, more than the determined fair share of layer-4-traffic messages by setting the weight assigned to the local computational node to 1.

9. The integrated, application-aware load-balancing component of claim 2 wherein the local layer-7 load-balancing subcomponent assigns target weights to each of the multiple computational nodes inversely proportionally to the number of layer-7 messages queued to the computational node for processing.

10. The integrated, application-aware load-balancing component of claim 9 wherein the local layer-7 load-balancing subcomponent assigns target weights to each of the multiple computational nodes proportionally to a number of layer-7 messages queued to the computational node for processing by:
computing an initial value as 1 divided by the number of computational nodes of the multiple computational nodes; and
for each

Assignees

• Vmware Inc

Inventors

• Agarwal Mayank
• Maderer Eyal
• Herlocker Jon
• Morales Ramsés

Classifications

• H04L69/326 Primary

  in the transport layer [OSI layer 4] (H04L69/16 takes precedence) · CPC title

• H04L67/1002 Primary

  Electricity · mapped topic

• H04L69/329

  in the application layer [OSI layer 7] · CPC title

• H04L47/125

  by balancing the load, e.g. traffic engineering · CPC title

• H04L67/1001 Primary

  for accessing one among a plurality of replicated servers · CPC title