> That is the real problem. I have to assume whenever you start implementing raw
> devices, you go far far away from DBMS design to OS design. Perhaps you do not
> have to write the process control part of an OS, but (I mean this jokingly) in
> one's arrogance one must think you can access the HD's more efficiently than the
> OS can. Hey, maybe some can do this, and have enough knowledge about HD's and
> controllers to do this, but making this portable is got to a _lot_ of work.
>
> Not that it couldn't be done, or that it wouldn't be eventually a good idea
> (*though I am not certain about that)...james
this reminds me so much of the multi threading issue Sybase faced. When they
were adding support for multi threading, every Unix has different level of
support that they end up added their own threading engine.
I for one, don't see RAW device support in Postgres has a "critial" missing
feature, in my thinking, there are other things that are more important.
Though for something like RDBMS, where each person's need could be
so different, it truly is YMMV.
