Last Updated: February 25, 2016
· kenlakoo
2a92ad89c8bd4e8b45edc8f363e18aa3 normal

Hello "Distributed Computing"

This is a quick digest of my study about "Distributed Computing" recently.


What is "Distributed Computing" ?

If you don't know what is "Distributed Computing", these are some simple cases that make you interested.

  • Case 1: Twitter: Why twitter can deliver tweets to my followers so fast?
  • Case 2: Instagram: Why Instagram can store so many images?

Reasons to use

Need for Speed:

  • Speed to overall response time by decrease the queue time
  • Decrease network latency due to route ### Need for stability:
  • Fault Tolerance by preventing single point of failure ### Need for scalability
  • Every computing node has its own limitation to serve. It can server 2K users but not 2B users.

Reasons not to use

  • Building prototype . Note: not so many users yet
  • The request can wait. E.g daily statistic reporting.
  • (I still feature it out) *


Say Thanks

1 Response
Add your response

Big pimpin

I have been doing a lot of research into distributed platforms/applications lately, as I am building an application that will need to take advantage of a distributed platform. While I haven't been able to put all of the below into practice simultaneously, I have tried each individually (as well as many others) to help decide what should (or shouldn't) be in my application.

In my opinion, the only way to be continuously successful with any distributed application is to implement the following:

  • <abbr title="Hierarchical Model–View–Controller">HMVC</abbr>, <abbr title="Presentation–Abstraction–Control">PAC</abbr>, or similar system architecture
  • Package management/build tools (e.g. NPM, Composer, Gulp, Grunt, etc.)
  • Great server & client side caching
  • <abbr title="Testing Driven Development">TDD</abbr> + Continuous Integration service
  • Easily/quickly scalable servers
  • Good in-code documentation
  • Use plenty of open source libraries

Without these, managing and extending a distributed application seems to just become too cumbersome. Beyond these things, there are plenty of "optional" extras (such as using a build language, semantic versioning, and/or Git) that will greatly help the process.

Also, while it may take ~1 second to send a page to the user from the time the request was sent, the machine will usually process the request and move on to the next after about 200 or 300 milliseconds; the rest of the time is spent jumping between servers/routers to get from the server to the user.

Of course for more complex applications, the process time can increase some. But if you are using a using a distributed application, this increase can usually be negated or at least minimized.

over 1 year ago ·