Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

That is the exact opposite of my experience. I try very hard to have the database as dumb as possible.

- Having mixed workloads on the database servers make it almost impossible to profile and tune. By definition, a database answers a lot of queries from a lot of clients. If these queries are hybrid data fetching / compute, then it's pretty much impossible to make sense of any system metrics.

- Having hybrid workloads usually lead to considering DB servers as black boxes (because you cannot make sense of their resource usage easily). This in turn leads to having hyper pumped up DB servers with both crazy RAM, crazy CPU, crazy caches, crazy SSDs and crazy HDDs.

- DB vendors of course understand the previous point, thus why Oracle switched to a "pay per core on the machine" model...

- The DB logic layer ecosystem is most of the time subpar compared to a more traditional programming language. If you depend on your DB server performing a lot of work, you will want to make these awesome queries accessible. That is, you will want stored procs, extensions, etc. Dealing with those is a huge pain to test, version, update...

- Overall all RDBMS have special SQL extensions all over the place. You're probably using tons of those without even knowing it. With time, vendor locking will become an immense burden and you'll be forced to pay millions to Oracle to keep your business running.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: