Tutorial: Implementing Facebook's BigPipe Using ASP.Net MVC - Part 1
José M. Pérez / September 18, 2010
4 min read • ––– views
Parts of the tutorial
- Introduction to BigPipe
- How ASP.Net MVC fits in the model. Registering and generating pagelets
- Browser implementation of BigPipe. Loading pagelets and their resources effectively
- Check out the demo Visual Studio solution
Through a series of posts I will explain how we can implement BigPipe Facebook using ASP.Net MVC.
In this first post I will describe what BigPipe is and sketch how we can make a similar implementation using ASP.Net MVC.
You can clone the code from the BigPipe project on GitHub that includes all the source code needed to run the sample.
BigPipe is a website performance technique used and coined by Facebook to serve web pages improving user's perceived load speed. In general, it consists of serving quickly the main content of the page, and then serve the content from other regions of the page called pagelets.
The implementation of these pagelets is performed in parallel on the server and served to the browser as soon as they are generated. This allows:
- Browser can start rendering the page content earlier (early flushing)
- Pagelets are served as soon as they are ready and the browser can render them in their container.
- If one pagelet takes longer to run, it will not delay the generation of the rest of pagelets.
- Pagelets are generated in several concurrent asynchronous threads and when a thread finishes its execution, it flushes the content so the browser can start rendering.
Sequence of the different stages during a pagelet generation:
Apart from the pagelet generation, parallelism is also applied during the processing of the pagelet by the browser. Each pagelet can define a set of CSS and JS files that it needs to work properly. These files are requested in such a way that it keeps a good performance.
Graph showing how pagelets resources are requested:
- For each pagelet, request in parallel every necessary CSS resource
- Once a pagelet has the necessary CSS files, inject the HTML code inside its container.
- When every pagelet has finished request its CSS files and is inserted in the document, proceed to request in parallel the JS files needed by the whole set of pagelets.
By keeping these steps we make sure that the pagelets are appended to the document and set their style, avoiding a FUOC (Flash Of Unstyled Content). By delaying the download of JS resources we are prioritizing CSS requests for content be shown earlier, as well as other pagelets to be downloaded.
- We take advantage of the parallelism in the server side, running simultaneously multiple pagelets.
- In general, improving the parallelism of the system. As the browser renders a region, some others are being transmitted through the wire and some other being built in the server.
- Everything is done in a single request by the client. You could implement BigPipe using Ajax calls from the browser to generate the pagelets, but at the expense of a greater number of requests (this will be covered in another post).
- When you flush each pagelet more packets are sent from server to client (however we can always decide to send more than one pagelet in each flushing).
- Browser has to interpret the code and insert pagelets in their containers. This causes repaintings and reflows which can be annoying to the user.
In part 2 of the tutorial I explain how to use ASP.Net MVC to implement BigPipe, using RenderActions and threads pool to execute the pagelets.