- From: Robert Hundt <robert.hundt@gmail.com>
- Date: Mon, 12 Nov 2012 11:31:57 -0800
- To: Tobie Langel <tobie@fb.com>
- Cc: "public-web-perf@w3.org" <public-web-perf@w3.org>
- Message-ID: <CAOWjH=CTrMqyqcKpdcP+q5MdOdzvQCL90SYzA-im6Q+Sy4kk0A@mail.gmail.com>
Thanks, Tobie, Needless to say, we (Gmail/Apps/others at Google) are really interested in this and welcome any help and participation. The slides are here: http://www.w3.org/2012/11/performance-workshop/agenda.html Best Robert On Fri, Nov 9, 2012 at 2:14 AM, Tobie Langel <tobie@fb.com> wrote: > Hi all, > > I saw Robert's slides on delta updates and read the workshop's minutes. > > I would like to add a couple of points as native delta updates are, with > AppCache, one of the keys to allow building rich client-side apps on the > mobile Web platform. > > A typical rich client-side application is easily composed of hundreds of > different JS files nowadays. In order to fight latency issues, developers > (rightfully) concatenate all of their JS files together to transport them > to the client. Admittedly, with SPDY/HTTP2 this might no longer be > necessary, but that still needs to be proven. > > While concatenating JS files is a net gain in terms of latency, it also > implies that a single line change in a single JavaScript file causes a > complete cache bust. This issues is orthogonal to whether or not AppCache > is being used. With continuous deployment techniques increasingly getting > traction (e.g. Twitter and Facebook push daily to their mobile website), > this implies daily cache-busts, which artificially inflates bandwidth > usage and increases load times. > > There are various techniques to fight this. Among the mobile web apps we > surveyed[1][2] as part of our fixing-appache effort[3], pretty much all of > them stored their JS in localStorage and did some form of delta updates on > it (whether it was true diffs or per module updates). But there are > performance associated costs with this (getting stuff from localStorage is > not free[4]), security concerns[5] and a large engineering overhead which > is paid for every app that wants to implement this. While WebCrypto APIs > help mitigate the security threat, they still imply an overhead to the > developer who needs to be aware of the threat and defend against it > accordingly. > > For these reason, diff updates should be transparent to client code and > probably belong at the http level. Some vendors have already expressed > interest in this. It would be fantastic to see it through. Suggestions as > to what the best platform for moving this forwards, and what stakeholders > besides vendors should be sitting at the table are more than welcome. > > Best, > > --tobie > > --- > [1]: https://etherpad.mozilla.org/appcache-london > [2]: https://etherpad.mozilla.org/appcache > [3]: http://www.w3.org/community/fixing-appcache/ > [4]: > http://www.nczonline.net/blog/2012/04/25/the-performance-of-localstorage-re > visited/ > [5]: > http://lists.w3.org/Archives/Public/public-webcrypto-comments/2012Aug/0057. > html > > -- Robert Hundt
Received on Tuesday, 13 November 2012 14:32:32 UTC