By collecting a lot of random writes in the BBU's writeback cache, and causing additional IOs for it's internal readahead, IO latencies grow absurdly. And on e.g. plenty LSI SAS controllers, the queue management for that is wholly unusable.
-
-
Show this thread
-
Note that for other workloads, with fewer random writes (e.g. increasing shared buffers to contain the whole working set, making small sequential writes more important by using logged tables in an OLTP setup) and more synchronous writes that's not true:pic.twitter.com/B8ru6zbsFf
Show this thread -
The queuing logic inside the controller still clearly is crappy, but it's outweighed by being able to acknowledge WAL flushes. So, WAL should be on a drive with the BBU's caching enabled, data absolutely not.
Show this thread
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.