Learning MCollective: Parallel Server Management in Puppet and Chef by Jo Rhett

Learning MCollective: Parallel Server Management in Puppet and Chef by Jo Rhett

Author:Jo Rhett [Rhett, Jo]
Language: eng
Format: epub, pdf
Tags: Machine Theory, Distributed Systems & Computing, Computers, Data Processing, Reference, Systems Architecture, General, Hardware, Computer Literacy, Computer Science, System Administration, Information Technology
ISBN: 9781491945711
Google: StxFBAAAQBAJ
Publisher: O'Reilly Media, Inc.
Published: 2014-08-13T21:01:49+00:00


Reaching Globally Diverse Servers

The first thing you should consider with globally diverse brokers, especially if the links are high-latency transcontinental links, is to localize as much traffic as possible using distinct collectives in each region. If you split up the nodes into local collectives and send most messages to the local collective, it will reduce cross-site network traffic significantly. We’ll show you how to do this in Chapter 12.

The second improvement is to set up your broker links in a star configuration if possible, or a modified star configuration if necessary. You want to minimize the hops (i.e., the number of brokers that have to process a request) between the client and server. This will improve the latency of the reply and avoid running into timeouts.

Timeouts in large networks are not always avoidable. Multicontinental networks create unavoidable high latency between your brokers. You will need to tune your requests to work around this:

One tunable is the discovery timeout. In a production environment with seven global sites, we found that five seconds was necessary for discovery to get answers from all nodes.

The other tunable is the request timeout. In the same environment, we found that the minimum suitable request timeout was the same number.



Download



Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.