Author: Jesper Louis Anderson.
Writing code intended to fail… improves performance? In this non-intuitive paradigm, chaotic fault injection forces developers to write more robust code. This results in a product that is more resistant to small faults and ultimately better performing.
Author: Anirudh Surendranath.
Here is a good discussion of the various factors that affect page speed. From perception of speed to importance of outliers, this discussion touches on the top aspects of page speed as well as improvements.
Author: Alex Browne.
Owning a Ruby on Rails blog site hosted on Heroku that was performing adequately was no reason to not improve performance! Alex Browne set to see how fast he could get on a tight budget. In all tests and metrics the new static site running on Jekyll on Octopress is much faster and capable than the old version. For only pennies a month, this is a consideration for anyone with a blog that is starting to cost too much.
Author: Ian Applegate.
This technical article explores TCP congestion operation and settings, ultimately creating a case for upgrading the linux kernel. Updates in the 2.6.38.x kernel and more in 3.2.x provide enhanced and additional algorithms to tune congestion-affecting settings. There is no one best fit solution, and an approach of monitoring and adjusting is advised.
Author: Peter Zaitsev.
Performance on a datacentre’s TCP network can sometimes be significantly worse than expected. TCP Throughput Collapse, also known as TCP Incast, affects many-to-one TCP links resulting in severe underutilisation of capacity. This is seen particularly in clusters where simultaneous requests are made of many nodes. While this article has a focus on erlang, anyone dealing with similar situations will be interested, particularly cluster storage, web search and MapReduce operations.
Authors: Raja Appuswamy, Christos Gkantsidis, Dushyanth Narayanan, Orion Hodson, and Antony Rowstron.
Conventional wisdom is that scaling out to a cluster is the way to grow into larger data. This paper by Microsoft researchers demonstrate that with realistic, sub-petabyte, operations it is in fact better to scale up to a larger server. Included are recommendations for Hadoop tuning for scaling up as it is designed to work for scale out. Their results imply that future software infrastructures need to be designed both for good scale up and scale out.
Author: Steve Souders.
If your company is new to ingraining web performance optimisation (WPO) in every department you may be reading this as the new “performance lead” in the company. This article covers all the bases in establishing a culture of performance within an organisation.
Author: Nick Galbreath.
By applying performance improvement tools used on external facing websites and products to intranet sites used internally, teams can become more productive and have a better experience at work. When thinking of customer satisfaction, don’t forget about employee satisfaction. The same rules and tools apply. “By creating a culture of data collecting and sharing, you will be able to better examine the data flows inside your organization, and find the cause of many unexpected problems.”