Sebastian, April 26, 2017 - 8:25 am UTC
I also was wondering about this.
Interaction with datafiles is done mostly with default 8kb blocks (sometimes 16kb, I rarely seen different configurations) while standard blocksize of redo log is 512 bytes so it is much lighter to write sequentialy 512 bytes of redo log and return command to session after it commit than write 8kb of datafile block (this is done via dbwn asynchronous to user session).
Although nowadays redo log can have 4 kb of blocksize which is reducing benefit of smaller writes.
IMHO I think that someday in future flash storage will dominate market so much that Oracle will have to redevelop RDBMS engine to elimante redo log entirely ;)
April 27, 2017 - 6:25 am UTC
These things do happen occasionally....Oracle 6 was pretty much a rewrite from the ground up
Reason
Asim, August 07, 2022 - 2:13 pm UTC
And what became the reason I mean what technology advancement became the reason for rewriting Oracle 6
August 08, 2022 - 1:22 am UTC
It was setting the foundations for pretty much all versions up to oracle 12 (multitenant)
Better rollback segments, space management etc. I think we can be pretty confident that it was not a decision taken lightly, because rewriting your product generally doesn't end well for companies. (eg Netscape)
Will online redo be removed in near future
Asim, August 08, 2022 - 4:53 pm UTC
So that was said in year 2017, now in 2022 (5years), do you think flash storage dominated the market so much ?
Do you think oracle will remove online redo log files and directly writes to data file, or atleast make the use of online redo log files optional in future to take the advantage of flash storage?
August 09, 2022 - 3:05 am UTC
If you're writing direcly to datafiles, now you have a contention issue - what if 2 people want to write to the same block?
With redo, because the writing is *always* append, ie, write *new* blocks, then you don't have this concern.
However, look at our Exadata stuff - we've been taking advantage of flash in various areas for many years. I would be unsurprised if we dont continue to exploit it in more ways
Concepts
emaN, August 10, 2022 - 12:40 am UTC
Do you think oracle will remove online redo log files and directly writes to data file, or atleast make the use of online redo log files optional in future to take the advantage of flash storage?
RAM is still faster than flash storage. Data and undo blocks changes are cached in memory. The redo log tracks changes in the sequential order. This protects data changes from inconsistency in the event of a write failure.