Comments on: Data Layer Diversity: It’s Not Just Relational Anymore

By: Scott Sanders

Scott Sanders — Tue, 31 Jan 2006 12:55:09 +0000

Sometimes an RDBMS is just plain overkill. FeedLounge does a very large number of key->value lookups, and the database was just overloaded with the writes.

Moving some of those reads/writes to a Berkeley DB backend uses almost no CPU, and the database is free for other work.

FeedLounge is also starting to use MogileFS to offload a large portion of the content processing.

Since our application is not just 'read mostly', the architecture has to get a bit more complex to deal with it. Tradeoffs for the greater good.

Note that FeedLounge has always and will always use a RDBMS for a portion of the system, but sometimes other storage/persistence models are more suited for the job.

By: Jaime Cardoso

Jaime Cardoso — Mon, 30 Jan 2006 13:30:08 +0000

Well the reasonings at the time were cost (Oracle and all were too expensive), Administering an LDAP server was a lot easier than adminestering a DB and, Netscape was interested in promoting LDAP (and had a lot of know how in developing over LDAP).
I don't think most of the reasons still hold today but, in a world where one size does not fit all, this just proves that there are other ways.

By: stephen o'grady

stephen o'grady — Mon, 30 Jan 2006 11:34:28 +0000

cote: i'd have to dig around, but most of the open source vendors have some embedded success stories. db4ojects, in particular, has had a substantial degree of success in that arena.

Jaime: interesting – what's the advantage of using LDAP in that fashion? the lightweight aspects?

By: Jaime Cardoso

Jaime Cardoso — Sat, 28 Jan 2006 17:25:33 +0000

Some years ago, Netscape launched a e-commerce store that was an Web server serving some HTML forms and, in the backend, an LDAP server (Netscape Directory Server 3.x).
I'm not saying that LDAP is the best protocol for it but, since those stores were in production in several sites, it proves that it can be done.

By: Cote'

Cote' — Sat, 28 Jan 2006 09:00:49 +0000

Do you have any info on MySQL and other OSS DB's being used as an "embeded" or bunbled DB in products? And how that scales?

The reason I ask is because many of the scaling success stories — "Flickr, Google, LiveJournal" — are SAAS-based apps where scaling can mean "throw more hardware at it. But, in the product case (when the DB is embeded/bundled), that's not so much the case.

As I'm sure you know, I'm not one of those "on ramp" people either: BUT, I haven't come across much about using the LAMP stack (or parts of it like MySQL, etc.) in "shrink-wrapped" products.