Oracle Offers Custom Intel Chips and Unanticipated Costs 97
jfruh (300774) writes "For some time, Intel has been offering custom-tweaked chips to big customers. While most of the companies that have taken them up on this offer, like Facebook and eBay, put the chips into servers meant for internal use, Oracle will now be selling systems running on custom Xeons directly to end users. Those customers need to be careful about how they configure those systems, though: in the new Oracle 12c, the in-memory database option, which costs $23,000 per processor, is turned on by default."
Re: (Score:1)
Fully agree. The two links cited in summary are totally unrelated to each other and the second link has nothing to do with the title of the story. The only commonality between the two is that both are anti-Oracle. Nice job Dell, HP, EMC.
Re: (Score:2)
Sales flow chart. (Score:5, Interesting)
Here is a flow chart to decide whether to buy Oracle products:
<Do you enjoy being utterly fucked over?> Yes--> Buy Oracle. No--> Run for the hills.
I've been at two places which have been Oracle'd. It's like being pwn3d except you end up $10,000,000 poorer. You also end up with less dignity than the inevitable tebagging you might get in Halo.
Re: (Score:3, Informative)
There are two RDBMS products [1] for the top end. One is Oracle, the other is DB/2. Neither is cheap.
Oracle is more tunable, DB/2 tends to "just work" for the most part. Of course, IBM can hand you a decent DB/2 stackin one package. Not cheap, but DB/2 running on a zSeries or a POWER7 is going to take some work to bring to its knees.
As for MS SQL, it is getting better. There are tasks where I'd never think of using it in the past where it can easily handle today.
[1]: We are meaning ones that have some
Re: (Score:2)
Re:Sales flow chart. (Score:5, Insightful)
DB/2? What about PostgreSQL?
Because PostgreSQL doesn't support shared-storage, active-active clusters. PostgreSQL "clusters" use replication to provide a warm standby using separate storage.
So you need twice the (high-speed) disk storage for a PostgreSQL solution.
That's just the database. Now you need to add clustering/HA to that, with pgpool. And pgpool is a turd. Yeah, it's better than NO standby/cluster. But set up a test PostgreSQL/pgpool cluster and really start beating on it - pull some plugs, shut down hardware, "kill -9" some database and/or pgpool processes. And watch pgpool piss all over itself.
In short, if you want a true clustered database solution, it's Oracle or DB2.
Re: (Score:1)
What about MySQL/MariaDB? Facebook doesn't even need a full time person for their production MySQL database cluster. It doesn't use shared storage, but if a shared storage system handles your workload, you're not all that big anymore.
Re: (Score:2)
How does PostGreSQL compare?
Re:Sales flow chart. (Score:5, Interesting)
How does PostGreSQL compare? Pretty well. I used to be an Oracle DBA (between Oracle 6 and 10g) but now much prefer Postgres. At the very high end of things, Oracle may well perform better. But Postgres is much better to work with, has excellent support organisations (unlike Oracle who will charge you a fortune to mostly just waste your time), is very feature-rich, and is generally a pleasure to use. If you have such data and transaction volumes that Postgres simple won't cut it, you should probably question whether Relational is the right paradigm.
Give Postgres a try, it's pretty easy to get started. And if anyone tells you MySQL is faster, ignore them until they prove it using your application and realistic transaction volumes.
Frankly I wouldn't touch Oracle with someone else's 10 foot pole.
Re: (Score:2)
While I agree with you about Oracle, you are dead wrong on MySQL. MySQL(well, increasingly MariaDB) is extensively used at some of the highest trafficked sites on the internet(Google and Facebook to name a few). It is more than capable of handling large transaction volumes.
Re: (Score:2)
In my experience MySQL can indeed handle large transaction volumes, but only with relatively simple queries.
Throw some complex joins in there across multiple tables and the performance plummets. Run multiple concurrent similar queries and you can find yourself getting exponentially worse performance.
PostgreSQL on the other hand may not have quite the same level of raw performance on simple queries as MySQL, but it tends to cope much better with more complex things.
Re: (Score:1)
How does PostGreSQL compare?
Poorly, I'm afraid, from a pure performance point of view. Think for instance heavy multi-join queries on a star schema where PG can't implement various optimization algorithms because PATENTS. And that's not even going into really enterprise-oriented optimizations, tools, support and so on.[*] PG is a truly awesome db, but to make the best of it you need to understand its strengths and limitations, as with everything.
[*] another part of the problem is stupid company policies concerning data partitioning an
Re:Sales flow chart. (Score:5, Informative)
I work at a large government department with stupidly large scientific datasets being thrown in and out of databases and we're migrating as fast as we can from Oracle to Postgres. The only thing we can't really shake is bloody Oracle financials and a few crufted old Java apps that we don't have the code to rewrite.
Postgres handles beautifully, and on some things even better although on some nasty multi-join type things Oracle will still beat it.
But it doesn't even matter because we can just throw more hardware at it infinitely cheaper than the extortion racket that Oracle pricing represents.
MariaDB is surprisingly competent too and in fact even has a surprisingly complete GIS implementation (Although PostGIS is the gold standard as far as we are concerned). Just avoid the Oracle branded one (MySQL), its not as well tuned, doesn't play nice with packaging systems and is generally posessed of the Oracle odour.
Re: (Score:1)
Re: (Score:3)
PostGres compares ok on a lot of workloads, but when the rubber really hits the road that is when it starts to fall apart.
They must fix the TXID ID problem. It will now at least shut down when it is getting close to rolling over, but the vacuum process will just kill your performance in very high transaction workloads. Not that Oracle would not have the same problem if they were using a 32 bit number for the value, but with the size of the ID Oracle uses this won't happen for ~ 140 years.
Immovability...
Re: (Score:1)
How big is big enough that nothing but Oracle will do? Facebook is on MySQL, Wikipedia is on MariaDB and Google is using Bigtable.
Re: (Score:2)
It's more the nature and size of access rather than the sheer volume of data. Facebook and Wikipedia both act on small portions of the overall dataset, Wikipedia additionally is a read-mostly workload, and Google's access patterns aren't suitable for a relational database.
Re: (Score:2)
I don't disagree with you, but I'll also add that there are some kinds of environments which need a huge DB like Oracle.
Because, let's face it, SQL server doesn't really scale up to the same level of performance, no matter what anybody tells you.
If you're big enough that nothing but Oracle will do, this is the cost of doing business.
If you believe SQL Server actually provides Enterprise class solutions ... well, you aren't very well informed. It simply doesn't handle stuff on the really big end of things.
Not saying that Oracle aren't greedy bastards who gouge their customers, but sometimes you really do need a bigger environment.
So doesn't Oracle's SPARC/Solaris line already cover this?
Sales flow chart. (Score:2)
Anyone who buys solutions deserved to be parted with their money.
Re:Sales flow chart. (Score:4, Insightful)
Here is a flow chart to decide whether to buy Oracle products:
<Do you enjoy being utterly fucked over?> Yes--> Buy Oracle. No--> Run for the hills.
I've been at two places which have been Oracle'd. It's like being pwn3d except you end up $10,000,000 poorer. You also end up with less dignity than the inevitable tebagging you might get in Halo.
I'd just like to confirm... the OP is not exagerating at all here. Oracle is today, what Microsoft was 10yrs ago.
They're big.
Their customers are currently trapped.
Oracles Management think that this situation will last forever and can't imagine a time when customers would move to something else.
They are using that power in such a drastic and barbaric way that, as painful as it may be, there's just no way they are going to continue using them in the future.
In 10yrs we'll all have moved on, and Oracle Execs will be scratching their heads wondering what happened to the gravy train. Just like MSFT is doing now.
$23k isn't crap to an oracle shop... (Score:1)
$23k is nothing but pennies to an oracle shop.
Posting anon as I'm a unix sysadmin in an oracle shop.
Re: (Score:1)
Oracle lives to audit. (Score:1)
Any perceived slight against them by a customer invites an audit of epic proportions.
My organization just wrote them a 7 figure check due to some sprawl in our environment that wasn't properly handled.
We did it proactively, through a partner, and held off the 8 figure check we might have had to write.
I'm not that guy, but I know his world.
Re: (Score:2)
So I used to be a DBA + sysadmin at an Oracle shop ~10 years ago.
Someone even managed to talk Oracle into selling us a site license for *everything* for $1mil/year. (a steep education discount; this was a university).
Unfortunately, they couldn't get the various schools and departments to agree to pool their money to buy the site license, so instead we paid more for restrictive licenses and were prone to auditing. The only reason I saw for not buying into the site license was if departments were planning o
Re:$23k isn't crap to an oracle shop... (Score:4, Interesting)
It wasn't so much a kickback, as an offer of a highly paid, no show job at Oracle after the contract closes.
At least that's what I've personally witnessed.
The company involved was under rate base, so they added 15% and passed it on to the electric ratepayers.
That said, Oracle financials? At least in the case above it was the DB. Everything else Oracle sells has _negative_ utility. You could get it done faster and more accurately with a yellow pad and slide rule.
Re: (Score:2)
$23k is nothing but pennies to an oracle shop.
Posting anon as I'm a unix sysadmin in an oracle shop.
Yes, but after becoming an Oracle shop, you don't have any pennies left to spend. And $23K per processor isn't really pennies to anyone. If you're spending the big bucks already, you have tons of processors. If you aren't, then it's massive.
But the real problem here is that it's done by default, regardless of if it's needed at all. So a client ends up spending that money, very likely on something they don't need and don't see any benefit from. Let's assume they have only two machines running Oracle, in
Re: (Score:2)
Good thing 'the guy' has a lucrative offer from Oracle...but in your case, unlikely. $100,000 isn't a lot of money to Oracle sales. Not enough for them to buy the purchasing agent.
Re: (Score:1)
I should point out that on multicore x86 machines Oracle counts 1 processor license needed per 2 processing cores. You're probably looking at 4-6+ CPU licenses for a dual processor system.
Re: (Score:2)
The 8895 is used in the Exadata Database Machine X4-8,an 8-processor rack system with up to 12 TB of system memory 672 terabytes of disk, 44 terabytes of high-performance PCI Flash, 240 database CPU cores, and 168 CPU cores in storage to accelerate data-intensive SQL.
The article implies it would be at least 8 processors (I hope they don't charge by CPU or CPU core). Anyway, it's at least $200k. But as you say, an Oracle shop is already in way deeper than that.
Re: (Score:2)
$23k/core pricing is stupid greedy. Cores are not getting much faster and therefore chip companies are adding more cores to increase performance. Oracle DB pricing should be constant per socket regardless of the number of cores and whether the CPU is implemented as a multichip module or single chip.
Re: (Score:2)
Posting anon as I'm a unix sysadmin in an oracle shop.
We'll try not to hold that against you :-)
23k or.. (Score:2)
The whole point of going in-memory inside the main 12c database is that sometimes the alternative to the $23k (list price; negotiate 60-90% off that) is buying a new CPU and licensing the whole database ( + options + OS + etc -> far far more than $23k) on that.
So although normally I bemoan Oracle's exceedingly unfriendly licensing model on this occasion it's not terribly surprising.
We are talking about Oracle customers (Score:5, Insightful)
Is that even a contract? (Score:2)
Expect this shit from Oracle (Score:1)
This is what I'd expect from Oracle's "Well, how much ya got?" mafia-style pricing.
Just turn a copy of your books to your sales contact Vinnie 'flat table' Malone and he'll let you know how much of your gross will be required to keep your data safe.
(Such a nice data warehouse you have there. It would be terrible if something.. Unfortunate would happen to it)
Re: (Score:2)
The link appears to be made in TFA (the first one):
So, what does the in-memory database option do? (Score:5, Interesting)
So, what does the in-memory database option do? (Score:1)
In the dark days of computing history before AJAX was even conceived and Mad Men were still crazies, "in-memory databases" meant that the database INDEX was in RAM (ideally if you DB admin was worth their salary), but then people wanted to pretend their were the next Google, famous for their massive search index in the pentabytes of storage, so hipsters started the NoSQL fad to be awkward like middle-aged men in skinny jeans as a vain attempt to self-proclaim their importance.
Now Oracle is making money by s
Re: (Score:2)
So, this Xeon model has special instruction set that helps keep the in-memory index synced across all cores in the server. Here is an Intel brief describing the technology and
Only 23,000? (Score:4, Insightful)
This is like pennies to someone that can afford to run Oracle on custom hardware. Why is this even newsworthy?
Re:Only 23,000? (Score:5, Interesting)
I was really surprised that Oracle did not build database optimization right into the M series SPARC chipset like SUN did for the T series and Java.
DB/2 on IBM hardware definitely gets a boost from software/hardware integration.
That's EllisON and he's a top democrat donor (Score:2)
His name US Ellison, not Ellis. He's one of the top money men for the democrat party. Look it up.
http://www.opensecrets.org/ind... [opensecrets.org]
Real CPU article here (Score:2)
http://www.datacenterdynamics.... [datacenterdynamics.com]
Cores vs speed changes (Score:2)
So, basically, the "tuning" is just giving them a way to trade active cores for speed, changing on-the-fly without restarting. More cores active, slower speed each. Less cores active, faster speed each.
Kinda nifty, I think. Not sure why it should be limited only to Oracle, though. Seems like a performance idea with broad appeal and utility.
Re: (Score:2)
So, basically, the "tuning" is just giving them a way to trade active cores for speed, changing on-the-fly without restarting. More cores active, slower speed each. Less cores active, faster speed each.
Kinda nifty, I think. Not sure why it should be limited only to Oracle, though. Seems like a performance idea with broad appeal and utility.
From where I sit, I can't tell what makes it ultimately different than "BLU" for DB/2. Although I don't have a whole lot of call to do column-level in-memory work anyway.
In-Memory is not "turned on by default"! (Score:3, Insightful)
Seriously, does anyone check their facts any more? By default it is turned off. You have to allocate some memory to the In-Memory Column Store by setting the INMEMORY_SIZE parameter and restarting the database. This is not going to happen by accident.
The parameter that is being discussed (INMEMORY_QUERY) which is enabled by default does nothing if no memory is allocated. You only get charged for the option if you turn it on by allocating the memory. This INMEMORY_QUERY parameter is not part of that issue.
Someone has taken something out of context and run with it. Now it has taken on a life of its own. Quality journalism!
Re: (Score:1)
Seriously, does anyone check their facts any more? By default it is turned off. You have to allocate some memory to the In-Memory Column Store by setting the INMEMORY_SIZE parameter and restarting the database. This is not going to happen by accident.
The parameter that is being discussed (INMEMORY_QUERY) which is enabled by default does nothing if no memory is allocated. You only get charged for the option if you turn it on by allocating the memory. This INMEMORY_QUERY parameter is not part of that issue.
Someone has taken something out of context and run with it. Now it has taken on a life of its own. Quality journalism!
Not to mention most people who spending the amount of money required for a custom built solution will already know the licensing requirements. There won't some magical surprise you owe us another $100k because you have 4 processors instead of 1. It sounds more like the OP can't read.
Re: (Score:2)
As ORacle controls MySQL, it might be free but for how long eh?
worth thinking about.
Re: (Score:1)
no thanks. Better, cheaper and far more flexible (Score:2)
Oracle's pricing is predatory nonsense. Anyone worth their salt has moved to MySQL, postgresql and most importantly NoSQL databases. Only old school IT is likely to put up with 23K per processor in today's multicore and highly distributed environment. And the last time I worked with Oracle RDBMS it still had a large number of the same warts I hated in their product way back in the 80s.
Just say NO!
I'm so sick of Oracle... (Score:1)