Skip to main content

Relational DataStores - How good?

An interesting updated by the twitter team about the scalability probs they are facing : http://dev.twitter.com/2008/05/twittering-about-architecture.html.

 

They say that twitter was not exactly architected as a messaging system but for a content management system. But users started using it to a messaging system. Another example of an application used differently from what it was initially designed. Previously there were social networking sites, which were intended initially to help people to connect, but then started being used to convey full blown information.

 

Anyway back to the topic. The twitter blog does mention some points worth considering. The queries they ran seem to be executing for too long. Ya, I know what they r talking about. The example they cite “consider 1000 persons following a user”, and the user updates her status multiple times. Whew…. Must be quite a challenge for the team at twitter.

 

So one thing I picked up is “stay away from relational databases for real loads of data. But then what is the solution? Need to look around first thing tomorrow morning.

 

 

 

Comments

Popular posts from this blog

B.O.F.H.

Was browsing the net, and found some old link to BOFH. Not sure as to how many used to read PCQuest some time back when it used to come, but i remember reading them, with of course re-reads and re-re,etc...i can go on you know. Anyway, it's about a system operator who is bugged by no-brainers asking silly computer related queries, and this guy repling with wierd answers. I had never known the origin of this character, until i saw this . Anyway for those who have never read BOFH, here are two of them.

So you have your website deployed in PROD ... now what ??

Posting on behalf of Usr.Web.Speed - My previous job had been to architect and develop websites for various customers. During that time my team and I have architected and developed various web applications mainly for enterprises. (But below info is not restricted to enterprises) Other than the usual development and testing tasks involved, our focus area was to abide by multiple SLAs. One of the primary SLAs was to provide the users of our websites a very low (usually subsecond) response time (or page load time). To adhere to this SLA, we did multiple activities, in code, process as well as infrastructure. These include (but not limit to) - Using best practices including (http://developer.yahoo.com/performance/rules.html) Determining the optimum number of calls to the databases, open connections, etc. Providing the fastest mechanisms to download associated content (such as stylesheets, JS files, etc. over CDN) And debugging the reason for the slowness of the websites, when ...