Thoughts on Exadata V3
I expect we’ll see some announcements at Oracle Open World in a couple of weeks (I don’t know if they will really call it V3 yet by the way).
DISCLAIMER: This is all complete conjecture on my part. I don’t work for Oracle and I have not had any conversations with any Oracle employees about official future directions or plans. (I probably would have had to sign a non-disclosure before I could have any of those discussions and then I wouldn’t be able to do this post).
text
Anyway, here’s the list of things I am thinking we might see:
Bigger Database Servers – I mean physically bigger, with open slots, so we can put HBA’s in them to attach to external storage for migration and backup purposes. Sun 4275’s perhaps since they are already using them as storage servers. The only real issue is that we’d run out of space on a full rack configuration, but if the machines are beefier, perhaps we wouldn’t need as many DB servers anyway.
Bigger Database Servers – I mean more memory and more processing power. Faster chips and bigger DIMMs are a no-brainer. Just put them in the existing 4170 boxes. But how about different models altogether. M series perhaps (which also means a change in O/S). Should be fairly easy to do actually as the DB already runs on Solaris. Might make the Sun shops really happy as well. 😉
More Options – I expect we’ll see a little more flexibility in the configurations, because “One Size Fits All” really doesn’t (or at least many people think that it doesn’t). Anyone want a 2/3 rack?
Incrementally Better Software – It’s a great leap forward already, but I expect more things to be offloaded to the storage layer (some of the analytic functions, some of the aggregate functions, etc…)
Exadata SAN – I could see Oracle announcing a stand alone storage unit with a variable number of “trays”. They might even announce some software for doing some of the more SAN like features (think Snap Mirror).
By the way, I doubt they’ll be buying Netapp just yet (need to drive the price of the stock down a bit first I think).
Well that’s what I’m thinking. What do you think?
Hi,
I don’t think Exadata V3 will be the big thing that gets announced on the OOW2010, It would get too boring – Exadata 3 years in a row.
I think it’s about time to finally release/announce the date for the Fusion Applications. It’s been talked about for several years already,
Maris
1. Cut down the price to benefit the SME market.
2. Disk-to-disk online backup like IBM flashcopy, EMC snapview.
3. 10g backward compatibility.
More like Exadata does Apps – really really fast .. So more stuff “certified” to run well on Exadata. I dont see any of your comments being big winners or likely events. Then again – neither is mine. Perhaps its the year of the Sun.
I really miss a synchronous replication (other than Dataguard) between 2 Exadata systems. Sync DG dramatically slows down heavy updates (I measured log file parallel write on the primary system = 7ms on a 10GbE link), not to mention the problems with logging/nologging operations. The technology to design an ASM mirroring between cells attached to two different sets of data base servers is there. This would allow streched clusters. Cisco Infiniband switches can have Ethernet and Fiber Channel gateways to extend the IB fabric(Sun IB switches do not seem to offer them though).
I’d like to see less distinction between a storage cell and a database node. Call it a database cell… Why do we need the distinction anyway? The Infiniband network is already used for both the interconnect and the “I/O”…. I’m sure it will be released as Oracle 12c (for “cloud-cell-cluster”)….
Dominic,
Ha. If we don’t separate the pieces people might start calling it the “A” word. I won’t even type it here for fear that Kevin and Dan might start sending me disparaging emails. You’re probably right about “cloud” entering the Oracle name space. They seem to really be liking that word right now. Kind of like “performant” was a year or two ago.
Kerry
[…] — Oracle has not made any official announcement – but blogger Kerry Osborne is speculating how the next version of Exadata will compare to V2. And, though much of the hype for next week’s Oracle OpenWorld conference has been around […]
[…] did a post a couple of weeks ago (Thoughts on Exadata V3) about what I thought we might get to see in the way of Exadata changes. We got most of the things […]
“note also the “enq: KO – fast object checkpoint” wait event which is done before the direct path reads (replaced by the “cell smart table scan” event in Exadata land).”
==> This is interesting and a serious replacement, if this is true. I tested on Exadata V2 machine when PX is disabled, wait event was “cell smart table scan”. When PX is enabled, each of the PX Slave was waiting for a short period time of “cell smart table scan”, but the QC was waiting much longer on “enq: KO – fast object checkpoint”. So what indeed does the the event “enq: KO – fast object checkpoint” mean in Exadata?
Yibin,
Not sure how this comment came about on this post, but I believe the “enq: KO – fast object checkpoint” represents time doing the same thing on Exadata as on any other platform. That being making sure all dirty blocks for the object are flushed to disk before starting a scan via direct path reads (or in the case of Exadata – Cell Smart Table Scan).