Hi everyone
I'm pleased to announce that Aaron Schulz is taking on a new role in
Wikimedia Foundation's Platform Team: Senior Performance Engineer.
Aaron works on MediaWiki internals -- components that every
user-visible feature depends on, but which are rarely user-visible
themselves. The impact of Aaron's work on the reliability and
performance of MediaWiki is felt by many, but fully known to few, so
I'd like to use this occasion to highlight some of this work.
Aaron is the primary author of MediaWiki's file handling backend, job
queue, its integration with Redis key-value database and the
OpenStack's Swift distributed filesystem -- just to name a few of the
more significant (and relatively recent) components. In addition to
these things, Aaron has designed a heavily-used set of generic
mechanisms for enabling concurrent access to shared resources, for
breaking up big computations into smaller units of work, and for
distributing those units of work across a cluster of machines that can
execute it. Aaron's code plays an important supporting role in most
(if not all) significant MediaWiki functionality.
Aaron's move to performance recognizes the critical role he has
already played in this area, and reflects our evolving understanding
of how important that work is. We are constantly striving to make the
experience of using MediaWiki snappy so that editors can do their work
without waiting on the software to acknowledge their actions. To meet
these standards, we need to tackle performance in (at least) two
areas:
1. Using data to provide a picture of how users experience the site,
identifying where users encounter latency, quantifying how it impacts
their engagement, and using this information to drive optimizations to
the code, paying particular attention to client-side network and
computational load. In making this a focal point, we are heeding Steve
Souders' Performance Golden Rule: "80-90% of the end-user response
time is spent on the frontend. Start there." [1] Our goal here is to
ensure that the visual display of information and the reactivity of
the user interface exhibit the kind of immediacy that is needed for
users to focus on and engage with content.
2. Building a well-integrated set of services that, when used in the
most obvious way, allows it to do as many things as possible, as fast
as possible, while remaining resilient to changes in site activity.
The vast majority of visits to our site never reach the backend of our
stack, but the minority that do involve really critical site
functions, such as all editor activity. We want the MediaWiki
ecosystem to have not only the capacity to sustain current levels of
activity, but to be an enabler of growth by making it possible for
many more people to contribute in new ways. To meet this challenge, we
need MediaWiki to be increasingly parallel, and to provide efficient,
concurrent access to a rich variety of content types and storage
layers.
This split maps neatly to Ori and Aaron's respective skill-sets and
experience, which makes for an effective partnership. Having Aaron
formally move to a performance role both recognizes the work Aaron is
already doing, and it lays the groundwork for a process for meeting
performance challenges that leverages their diversity of expertise.
Please join me in thanking Aaron for taking on this role!
Rob
[1] Steve Souders is formerly Head Performance Engineer at Google and
"Chief Performance Yahoo!" at Yahoo. Performance Golden Rule:
http://www.stevesouders.com/blog/2012/02/10/the-performance-golden-rule/