Comments on: A New DB for 80% of Facebook, YouTube-scale Sites http://www.roadtofailure.com/2009/08/07/a-new-db-for-80-of-facebook-youtube-scale-sites/ Scalability, Startups, and Computer Science. Thu, 26 Jan 2012 18:04:34 +0000 http://wordpress.org/?v=2.9.2 hourly 1 By: chris http://www.roadtofailure.com/2009/08/07/a-new-db-for-80-of-facebook-youtube-scale-sites/comment-page-1/#comment-132 chris Fri, 14 Aug 2009 15:23:41 +0000 http://www.roadtofailure.com/?p=58#comment-132 >> does it really matter that it’s “all the millions”? Maybe not. Google uses a lossy operation, as anyone who has ever tried to use it to do any kind of deep specific search will tell you. Try going down to page 15. >> does it really matter that it’s “all the millions”? Maybe not.

Google uses a lossy operation, as anyone who has ever tried to use it to do any kind of deep specific search will tell you. Try going down to page 15.

]]> By: David http://www.roadtofailure.com/2009/08/07/a-new-db-for-80-of-facebook-youtube-scale-sites/comment-page-1/#comment-117 David Wed, 12 Aug 2009 15:33:59 +0000 http://www.roadtofailure.com/?p=58#comment-117 Lossy database operations is an interesting concept. One of the invisible but limiting rules of the existing paradigm is that all operations must be 100% complete and perfect. But if I'm searching Google for web pages and there are millions of matches, does it really matter that it's "all the millions"? Maybe not. Lossy database operations is an interesting concept. One of the invisible but limiting rules of the existing paradigm is that all operations must be 100% complete and perfect. But if I’m searching Google for web pages and there are millions of matches, does it really matter that it’s “all the millions”? Maybe not.

]]> By: Bradford http://www.roadtofailure.com/2009/08/07/a-new-db-for-80-of-facebook-youtube-scale-sites/comment-page-1/#comment-95 Bradford Fri, 07 Aug 2009 20:25:48 +0000 http://www.roadtofailure.com/?p=58#comment-95 Yes, we certainly do get data locality -- it's very important, which is why we're doing our on-the-fly aggregation on the regionservers first, then sending it to the master node. If you'd like to tell me more about what your company needs so I can envision the product better, feel free to drop me an e-mail! We can chat on AIM if you'd like. Yes, we certainly do get data locality — it’s very important, which is why we’re doing our on-the-fly aggregation on the regionservers first, then sending it to the master node.

If you’d like to tell me more about what your company needs so I can envision the product better, feel free to drop me an e-mail! We can chat on AIM if you’d like.

]]> By: Robert Boyd http://www.roadtofailure.com/2009/08/07/a-new-db-for-80-of-facebook-youtube-scale-sites/comment-page-1/#comment-94 Robert Boyd Fri, 07 Aug 2009 19:17:03 +0000 http://www.roadtofailure.com/?p=58#comment-94 I thought this was the concept of Hadoop data locality -- the computations run on the machine where the data is stored. I thought you get this for free if you run Hadoop on the same servers running HBase, is that not so? On a side note, you say the operations are "communicative" and associative where I think you mean "commutative". In all this is an excellent post and exactly what our organization is looking for. Can we have it yesterday? I thought this was the concept of Hadoop data locality — the computations run on the machine where the data is stored. I thought you get this for free if you run Hadoop on the same servers running HBase, is that not so?

On a side note, you say the operations are “communicative” and associative where I think you mean “commutative”.

In all this is an excellent post and exactly what our organization is looking for. Can we have it yesterday?

]]>