Unconference
If you have a suggestion for an Unconference session, please describe it below, separating it with a header. If you'd like to make a presentation, please note that.
We should put some signatures on some PGP keys.
Please add your pgp key to PGP-keys and bring government-issued photo-id to the PGP signing.
I'd like to discuss the following points during an unconference session, I'll try to bring a few elements, not necessarily slides but at least a bit of data. It could be nice at the end of the first day given that it could complement PHK's presentation of intermediaries:
- how load balancers make the web scalable today
- how bandwidth has been growing over the last decade
- the challenges faced at each OSI layer
- what are the absolute minimal routing keys a load balancer needs to make it possible to design a scalable infrastructure
- how to aim for 100+ Gbps by processing these keys at minimal cost
I would like to discuss web page performance in high latency environments such as satellite networks. In particular, how to maximize parallelism and how to take advantage of Server Push.
Independent from the question of whether or not proxies are still needed, I would like to discuss the technical challenges associated with implementing a secure proxy solution. This includes:
- Being able to trust the proxy to be who it says it is and do the right things.
- Being able to select which pages (objects?) are accessible to the proxy and which must bypass or pass transparently through the proxy. Another way to look at it is be able to determine for which pages (objects?) privacy protection is sufficient versus stronger protection is required.
- Securing (with at least privacy) communication between the client and proxy. Is this always needed or can it be optional (up to the user) e.g. when the client and proxy are adjacent inside of a secure environment?
- Securing (with at least privacy) communication between the proxy and the server.
- Supporting cascading and/or parallel proxies. (Cascading seems really hard re user opt in.)
Of course, I expect some (if not most or all) of this stuff to be covered by existing agenda items. So, I was thinking Wednesday rather than Monday.
This 5-minute presentation explains how you can write a very fast HTTP parser.
PicoHTTPParser (which is used by H2O) is a HTTP parser 10x faster than http-parser (which is based on Nginx's parser) of Node.js that was around in 2014.
HTTP/1 parsing might seem like an out-dated topic, but in HTTP/2 we need to parse headers anyways, therefore sharing the implementation techniques of text parsers would be beneficial.
