The Migration from Erlang to OTP A case
- Slides: 13
The Migration from Erlang to OTP A case study of a heavy duty TCP/IP client/server application Francesco Cesarini francesco@erlang-consulting. com Mickaël Rémond mickael. remond@erlang-fr. org
The Product. . . n IDEALX runs in-house customer projects – Based on Open Source components – develop what is not available – release reusable applications on their site n Develop an Instant Messaging Server – – – For one of the largest French ISPs Handle 10, 000 simultaneous users Based on the Jabber protocol Scaleable during runtime Allow flexible addition of services
The Product. . .
Building the Prototype. n n A team of 4 software engineers Erlang Knowledge – All self-taught – Had been active in in-house Erlang projects – Used the erlang-questions mailing list n They had a working prototype in 5 weeks – Handled 900 simultaneous users – Had performance and reliability problems n Delivered to an impressed customer
Building the Prototype. n Decide to take in external help – Code & Architecture review – Improve performance – Explain OTP behaviours n The code review showed – Well written code – Little use of higher order functions – Lots of unnecessary concurrency – NO OTP behaviours or design principles!!
Building the Prototype. sockets multiplexing de-multiplexing state/error handling de-multiplexing sockets listener users
How could they miss OTP? How do I apply object oriented design in Erlang? Mickaël Rémond Joe Armstrong You do not need to apply OO design as Erlang has something more powerful called behaviours!
How could they miss OTP? n The name Open Telecom Platform – They were developing XML based products – They were developing products for ISPs – They were not developing telecom products n They did not find the documentation – Relations between behaviours not evident – No tutorials n Found very few OTP examples online
To the Rescue! n OTP training: 2. 5 days – Only behaviours were covered n n Code review: 1 day New system architecture design: 0. 5 days – No documents necessary n ‘Code rewrite’ examples: 1 day – Using higher order functions – How to remove deeply nested cases – Bit syntax, etc.
To the Rescue! sockets supervisor multiplexing de-multiplexing state/error handling simple 1 -1 de-multiplexing sockets listener users
To the Rescue! n 25 days work from prototype to product – Included training, redesign, rewriting, testing n n Code reduction of 50% Free goodies from OTP – Restart strategies, application control, etc. n Problems – Limit of TCP connections per Unix process – One listen socket for many Erlang nodes – Windows NT database performance
The Moral of the Story. n OTP design principles and behaviours fill a gap that existed in pure Erlang – Structuring of programs – Reuse of Code – Methodology – Development strategy n For some one downloading Erlang/OTP off the web, finding/understanding design principles and behaviours is not obvious/easy.
The Moral of the Story. n We who work with OTP should promote it better – Advantages – The gap it fills – More and better examples – Help existing open source projects n Hopefully, this will result in others not doing the same mistake – Happy users = Many users – Many users = Fun user conferences
