Browse Prior Art Database

DETERMINE DYNAMIC LINK DEBOUNCE TIME BETWEEN PEERS FOR DATA CENTER SWITCHING

IP.com Disclosure Number: IPCOM000245959D
Publication Date: 2016-Apr-20
Document File: 7 page(s) / 163K

Publishing Venue

The IP.com Prior Art Database

Related People

Natarajan Manthiramoorthy: AUTHOR [+6]

Abstract

A solution is presented herein to learn the link up time for each link up on a data center switching network. A network management system stores learned information in a table with switch-A-Model/Software Version, Switch-B-Model/Software Version, port-speed, transceiver-type, and link up time. This learning is used to tune the link debounce time for the same link types (same link partners, same port-speed, same transceiver-type links). This helps in ensuring that each link in the network has the benefit of link debounce as well as that the convergence time is optimal for real failures. In addition, by monitoring link down->up->down syslog timestamps, the optimal link uptime is determined to avoid a link from exhibiting synchronous link flaps due to mismatch of software linkscan on one side and hardware linkscan on the other.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 33% of the total text.

Page 01 of 7

DETERMINE DYNAMIC LINK DEBOUNCE TIME BETWEEN PEERS FOR DATA CENTER SWITCHING

    AUTHORS: Natarajan Manthiramoorthy Venkatesh Srinivasan

Swami Narayanan Anand Singh Ambrish Mehta

Anulekha Chodey

CISCO SYSTEMS, INC.

ABSTRACT

    A solution is presented herein to learn the link up time for each link up on a data center switching network. A network management system stores learned information in a table with switch-A-Model/Software Version, Switch-B-Model/Software Version, port- speed, transceiver-type, and link up time. This learning is used to tune the link debounce time for the same link types (same link partners, same port-speed, same transceiver-type links). This helps in ensuring that each link in the network has the benefit of link debounce as well as that the convergence time is optimal for real failures. In addition, by monitoring link down->up->down syslog timestamps, the optimal link uptime is determined to avoid a link from exhibiting synchronous link flaps due to mismatch of software linkscan on one side and hardware linkscan on the other.

DETAILED DESCRIPTION

     The port debounce time in a data center switching environment is the amount of time that an interface waits to notify the upper layers that a link is going down. During this time, the interface waits to see if the link comes back up, and if the link comes back up, upper layers do not even know that the link momentarily went down and came back up. This helps to avoid upper layer re-convergence for every momentary blip/flap in the link. With the increasing number of mixed vendor and mixed Application Specific Integrated Circuit (ASIC) products in a data center environment, there is no single default debounce timer that fits all the requirements. Choosing a conservative higher value as a

Copyright 2016 Cisco Systems, Inc.
1


Page 02 of 7

default debounce timer results in sub-optimal convergence in the event of a failure. Choosing an aggressive lower value as a default value results in loss of debounce functionality itself.

    Customer support calls for link interoperability issues together with the customer determining the optimal debounce through trial and error experiments is a very common activity. This is costly, both in time and expense, to both the customer and the equipment vendor. Moreover, when an optimal value is determined for the debounce, subsequent hardware or software upgrades require the same exercise need to be re-done.

Figure 1-Example Data Center Topology

    Figure 1 above illustrates an example topology for a data center network. Again, typical data center networks have mixed vendor and mixed ASIC products. There is no single default link debounce timer that works for all possible combinations. Choosing a

Copyright 2016 Cisco Systems, Inc.

2


Page 03 of 7

higher value results in sub-optimal convergence in the case of a real failure. Choosing a lower value results in lack of debounce entirely.

    For example, consider a switch connected to two different switches in a...