"They will, say, put compute on AWS and storage somewhere else."
That's now how enterprises that can afford Oracle products work. They also do not normally buy computing services from an online bookstore.
Larry Ellison laughed off the cloud seven years ago. If anything, cloud meant nothing more than taking out a new full-page ad on the back of The Economist. Nothing would change for Oracle, he laughed. Nothing. Well, it’s changing now, with Oracle’s people digging fast to extract their firm from the hole their captain's hubris …
AWS is not Amazon books... but putting "computing" in one cloud and storage on another, looks a recipe for a lot of problems to me - unless the two clouds have very high speed links (or storage is true icy ones, towards 0°K, mostly for mandatory archival reasons). Making the two systems communicate easoòy and quickly would be a problem of yours, not the cloud providers'. IMHO many companie would like to go with a single supplier - after all true, proper "business continuity" even within a single supplier should be one of the cloud advantages, isn't it?
I could see some cases when relying on multiper providers could make sense, but I guess most of the time it's better to go with a single one (if prices are comparable) - if Oracle will be one of them only future will tell...
put compute on AWS and storage somewhere else.
I'm afraid you lost all credibility with that statement. You clearly have no understanding of how most real-world applications work.
The reason why we have Hadoop, In-memory databases, direct attach Flash, infiniband attached storage etc is to reduce the time for data to move from storage to compute. Increasingly even 8Gb fibre channel between adjacent boxes is too slow, let alone between data halls, let alone between data centres. We make huge efforts to take the algorithms to the data, not the other way around.
And you want to put compute and storage in entirely different companies, security domains and probably different continents over the public internet, connected by RESTful services?
How about ditching the 90s era concept of hard allocations for resources and provision cloud resources based on resource pools? Give me W x Ghz X x Memory Y x IOPS and Z x disk space and let me carve it up into however I want (any number of CPUs or allocations dynamic etc). Bonus points if you charge for only what capacity is used.
Like many enterprises have been doing for a decade now with virtual infrastructure in their own data centers.
Yeah, not holding my breath. public cloud, ugh.