we have seen in our last couple of installations on Sparc Supecluster T4, Exadata X3-2 and also when oracle has patched and upgraded Exadata X2-2 they have not implemented Exadata smart flashcache writeback. This despite the fact that during the X3-2 launch at openworld this feature was attributed to providing a 30% advantage on x3-2 and a 10% perfromance gain on X2-2
Why hasn't this been added to one command so that it automatically puts it in writeback mode. Are there any downsides in doing so ?
I was under the impression that there are no downsides. Can anyone throw some light on this. I checked with our oracle contacts and they said they did not find any downsides on the oracle internal experts mailing list also.
I would like a confirmation of this so I can convince some of my more apprehensive clients to make the change.
P.S for those subscribed to the MOS exadata community sorry for cross posting ,but this seems to be more active...
Seeing notes like MOS #1557421.1 make me nervous about putting it in place just yet. It's one of those things where if it goes bad, it goes really bad. Unless you have a workload that is really bottlenecked on writes, I don't know that you'll see too much improvement by enabling it. For example, in one of the first Oracle presentations I saw announcing it, they made one slide to talk about the feature, then 3 slides about why you probably didn't need it. That said, I could just be paranoid.
Hi Andy, the fix for that particular issue has been out for over three months now. We have been running writeback in our test servers for 2 months. There should be benefits for OLTP if all reads and writes go to flash our testing shows cerain EBS batches are satisfied 99% from flash...
I know you administer at least 60 clients are none of them using writeback ?
It's been out since January, but I'm more worried about the bugs that I don't know about. Also, the fact that it happens to heavy write workloads (the ones that it's supposed to help the most) make me nervous. We do have some customers using writeback for their flashcache, but I'd say they're still few and far between. We've been doing quite a bit of testing in our lab, though, and I would expect that the next storage server patch release will include additional bugfixes to make it more production-worthy. At that point, I expect to start recommending it more than I am right now.
After patching to 126.96.36.199.1 we tested the writeback setting early May.
A gain between 10% to 30% was seen in the tests (X2-2 HR) and went for it in production (X2-2 FR).
The gain in prod was around 10% on average.
Both systems run without any issue even during high load intervals.
To answer your question.
As I understand it this setting was introduced on 188.8.131.52.0 but had a stability issue under high load.
So probably this is why it is (not yet) a default setting.