Heh, this morning was fun. Helped a friend make their DB 50% faster by reducing calls to fsync. Makes me wonder tho; are there any surveys of how/when to use fsync? I feel it differs a lot for offline/online workloads, especially when considering transactions and the like.
Ah yeah, O_DIRECT sounds exactly like what we need here. Dang, keep coming back to this, haha.
-
-
That also has a lot of tradeoffs since you also have to perform all io at aligned byte boundaries. So you have to back up reads and be careful not to zero previously written data with writes starting half way through a block. And you give up the os pagecache
-
But it's more of a "pay for what you use abstraction" than the normal file interface :] But it turns out that people actually want a lot of buffered IO features without realizing it. So many tradeoffs that people mostly are fine ignoring. Most file usage is incorrect OTOH
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.