Re: AWS Cloud Write IOPS
Posted by: Rick James
Date: July 04, 2015 11:20AM

Writes (eventually) must go to disk. Even so, the 400/sec may come from more than (or less) than 400 rows inserted/modified. This because of caching on the write side.

Reads may be cached in the buffer pool. So there could be 320728 rows read per second (mostly from cache), but only 10 needed to be fetched from disk.

320728 is under the heading "row operations". Reading a row involves:

* Locate, in the buffer_pool, the block containing the row.
* If the block is not cached, read it from disk (This involves an IOP)

That is, "reading a row" and "fetching a block from disk" are are opposite ends of the spectrum of things that go on.

Furthermore, there are writes that do not match 1:1 with inserts/updates: rollback/undo log, doublewrite buffer, index updates, etc.

Inserting 1 row in a table with 10 indexes will involve 10 writes to disk (minus any that can be cached with other writes). That's just for the leaf nodes; there could be writes for the non-leaf nodes in BTrees.

Options: ReplyQuote


Subject
Written By
Posted
June 24, 2015 04:12AM
Re: AWS Cloud Write IOPS
July 04, 2015 11:20AM


Sorry, you can't reply to this topic. It has been closed.

Content reproduced on this site is the property of the respective copyright holders. It is not reviewed in advance by Oracle and does not necessarily represent the opinion of Oracle or any other party.