Hacker News new | ask | show | jobs
by jnevill 3756 days ago
I'd like to know more about this too. Data storage and retrieval is such an important and integral part of an RDBMS and the switch to Linux suggests that they've had to tear up a lot of their low-level logic. How well they've done optimizing storage for EXT3/EXT4 seems to be pretty important. It would be interesting to see what's changed and how they managed this.
1 comments
oblio 3756 days ago
If past Microsoft is any reference, most of the software is cross platform. It is designed from the start to be cross platform. It's just that they never bothered to actually port it to anything other than Windows.

Of course, bits and pieces might be tied to a platform and the dependency chain might drag other parts down, but overall Windows, Office, etc. generally have some sort of abstraction at the lower level that could theoretically allow porting to any platform.

link
jnevill 3756 days ago
That's my concern though. When you look at other enterprise-sized RDBMS you find tight integration at the "machine" level. Oracle's storage, for instance, uses "Blocks" which are sized based on the block size of the disk they are writing to. This tight integration at such a low level is part of what makes these RDBMS's fast. I would imagine that instead of running on an abstract layer that can be commonly built across many OS's that an RDBMS like Sql Server would be more tightly integrated with the machine they are running on in order to maximize performance and disk space. If so, I would really love to dig through what that means for SQL server.
link
telotortium 3756 days ago
Might that not imply that the implementation is relatively little tied to a particular OS? Taking your example, as long as there's a way to get the block size for a disk, there's not much more OS-specific code to deal with -- in particular, adapting your code to the file system doesn't need to be done since the DB will essentially create its own anyway. Similarly, only a few threading primitives need to be used -- the DB isn't going to tie itself heavily to the vagaries of the async IO implementation, which does tend to differ heavily and pervasively between platforms.
link