Transaction File implementation question

[Greg Young]

Been reading through tons of old threads about varying transaction
file implementations. As many of the discussions are old (some over a
decade) what is considered the best way of implementing today? It
appears that o_direct is frowned upon, is the current preferred
mechanism to use mmap and basically append pages + partial sync when
writing (one issue with the second is that it doesn't work in say
windows where as o_direct does)?

We currently follow the added constraints mentioned by Linus in an old
email (lost link) in that we use o_direct but prealloc and only use
o_direct when reading. Is it worth looking at a mmap implementation?

Thanks in advance,

Greg

-- 
Studying for the Turing test
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/