Oracle was vague about when its new Sparc T series servers using the 16-core "Rainbow Falls" Sparc T3 processors were coming to market when they were launched a month ago, but two of the four machines are now shipping. And Oracle has slashed prices and boosted performance to show IBM and HP that it is indeed back in the Unix …
Entry level is: "16-core processor running at 1.65GHz with 16GB of memory and a single 300GB disk"
A single disk? Not a mirror? Oodles of processor power, oodles of RAM, Solaris, but only one disk to run it from? Surely no disk (boot from SAN) and dual disk for mirroring would have been more sensible options otherwise why bother with a rock solid OS when it's not running on redundant disks?
The licensing factor
For the purposes of Oracle licensing the cores on the T3 each represent a 0.25 multiplier. This, combined with the lower OPN list prices (vs SPA list prices) make the T3 based systems a much more cost-effective solution once you look at the larger picture.
Oracle on the T* boxes
We've had nothing but problems with running Oracle on our T* boxes. The M* series is much, much more suited for it and find the performance of the T* boxes inadequate for what we need it to do. Fewer CPU's with more ooomph per core is much better than lots of slower threads for 99% of all the database instances we have (both OLTP and data warehouse ).
If you do a search on the web our experience with T series and Oracle seems to be the norm rather the oddity. (there's a reason licensing is so cheap for Oracle on the threads... it generally sucks)
Re: Oracle on the T* boxes
I had exactly the same experience. Many slow threads are not good for any database workload. It might work for benchmarks, but not for real-life apps when you need to run a batch on limited (hot) amount of data. Concurrency is causing most of those slow threads to wait. Even old UltraSparcIII were faster for most database workloads.
It's horses for courses - I've seen many instances where CMT has worked well, and several where it hasn't. In general terms, the larger the number of concurrent active users, the better it works: contrariwise, if you have a small number of users, or a large number with a small number of database intensive actions, you'll often find straight line speed is more important.
This paper gives some good general tips:
Love the description
Fantastic potential names for the company. Spot on.
how when the writer produces "articles" on IBM, there's never any slant.
But Oracle / Sun? He's always trying (and usually failing) to get a dig in.
I guess one of the companies didn't give him the job he went for.
"...no one answers a question at Oracle. (Which means that the company's name should not really be Oracle, now should it? Perhaps Denial, or Hubris, or Mine Mine Mine would be a better name?)"
Put this on a protest sign and march back-and-forth in front of their headquarters and you could make YouTube another million $$$.
Total Cost of Acquisition?
Hmmm conspicuous by its absence from the prices and options in the Oracle store is the cost of support. These costs do look nice and low, but you have to wonder what Oracle bang on top of this for support costs, given the strangled cries heard from many Sun customers recently, I'd imagine its pretty significant
Annual support costs are 12% of customer purchase price, this covers hardware and operating systems. This is the same for all Oracle systems.
So they're comparing the chips to black tea now?
Perhaps Denial, or Hubris, or Mine Mine Mine would be a better name?
"No kitty you can't have any that's my pot pie!"
Fraser: A single disk? Not a mirror?
The picture of the new entry level server is worth a thousand words.
Sure, you can get a mirror, but you can go from 1 to 16 disks with this entry level model.
Some production deployments do not even use internal disks, booting off of a SAN, NAS, or external disk trays is pretty useful.
It all depends on your business needs are - one size does not fit all.