We gave SQL Server 2012 one year to prove itself: What happened? I reviewed SQL Server 2012, codenamed Denali, just over a year back and highlighted the major improvements in Microsoft's relational database. After a year in production, was I right, or have other features proved more important in practice? I'm bound to agree with my last review, so I polled colleagues who also work with SQL …
As an aside, I'll wager whoever coined the term "deep dive" wasn't actually a diver themselves. When you dive deep, you make a plan in advance, to the minute, and then you execute it. If you find the lost treasure of Blackbeard but you have hit the limit, then you execute your ascent plan, and come back tomorrow. Divers who don't do this end up becoming a permanent part of the site. It's a product evaluation; what's wrong with calling it that?
I think "deep dive" refers to jumping off the proverbial cliff.
For example:
Normal Dive: Jumping off the side of the swimming pool going under the water a little over your head.
Deep Dive: Jumping off of something high as in the cliff mentioned previously and going in well over your head. Well unless you pancake it but that's going to hurt.
Scuba Diving: Totally different ball game and probably started after the term was coined? (still good fun though)
Well, it should be kept under wraps a bit but it's going to be a huge feature in the upcoming version of SQL server. Then it will fully support several stages of fuzzy logic. All in a wild attempt to bring a more natural and reliable feel to the SQL server.
The beauty is that it will far exceed the previous comments of "it will work, or not...", with the new fuzzy logic module enabled you can be sure that it might work, or not. But there isn't a 50-50 percent chance of things failing or succeeding; the chances go MUCH deeper.
Although the module is best used on the Enterprise management level because these people know what they're talking about when it comes to chance calculations, it's also perfectly usable for home servers. Now you can finally have the best excuse as to why your website isn't working all the time: "Because the evil Microsoft enforced fuzzy logic on my SQL server!".
No one will believe you, but you have to agree that it sounds cool!
What you're experiencing here is merely the first stage of the plan.
Using our DB is a real example, then. We collect large amounts of data, much of it questionable quality, and do what cleaning we can (there's ever the whiff of the Augean stables about our DB). We derive saleable IP by analysing the data in aggregate, not in detail. If we needed a mirror then we might well be happy to lose a few thousand, or tens of thousands, of recent entries if it kept us seamlessly up. We wouldn't much care, nor would our customers who look at patterns in the days or weeks or longer.
@TheOtherHobbes
Suppose Microsoft Airlines could fly at 1/10th the cost of other airlines but still reasonably reliably. Not suitable for people but could be fine for cargo.
In our domain that's fine. Lots of kinds of data don't need 100% accuracy. I think claiming it does is what marketeers do to sell vast data crunching kit and if people allow themselves to be ripped off, it's their call.
"one missing row or wrong value may be considered unimportant"
Really this should be "one missing row or wrong value may be considered relatively unimportant" but relative is well ... relative. You wouldn't know how important it is without comparing the results to same analysis from the full data set. Pointless.
When performing statistical analysis on large data sets there no need to compare data sets to workout how important that one row was. The only answer is; it wasn't important
If it is one row in a hundred thousand or a million it is statistically not important
If you think that one row in a million is statistically important then you're doing your stats wrong
As soon as you need the max() or min() observed price for a given predicate, one row can easily be the difference between a meaningful result and fucking so far off that the exercise was pointless.
Don't trust people when they say "it doesn't matter" about accuracy, integrity etc. They are bullshitting you every time.
ps: I know what a standard deviation is, but it does not alter my point!
Proves my point
In large data sets (population samples) 1 data point should not be able to skew your results that much. In most uncleansed data sets, max and min are junk values anyway and their only use is in finding anomalies in the data. What you really want to find is the percentiles.
Did nobody read the bit where I said "Obviously, this is not always the case!" ?
Of course it depends on the data set but I've found a lot of cases where the latest data and even slightly corrupt data can be dealt with and leave the resulting reports with meaningful data. I would have thought that obvious enough not to have to emphasise. While I'm here I will point out that this kind of thing has to be done with the knowledge of the users so they know what might be missing and what that means for their reports.
Congratulations on having no idea what statistical analysis is.
What you are querying is an absolute question with an absolute answer. Something which by the sounds of things you would need to query a live system for as you are so clear in wanting the most correct answer (Any backup is out of date as soon as it's finished building)
Please please please don't apply for a job with me until you get a clue
Wow, a whole article about SQLServer2012 (a database management system) and not a word about the core DBMS functionality, just a lot of waffle about peripheral stuff that might be useful in some shops but pointless in others.
This article simply reinforces what I have always believed about the SQLServer "experts" pool, and that is, that for the most part they really don't know about the dbms engine or why it is terminally brain dead in vast number of ways, and apparently neither do they care.
philip
Not changed? Is mature? Try again on both counts.
<http://msdn.microsoft.com/en-us/library/cc645577.aspx>, look for geographical data types if you consider that core db, but in particular 'OVER Clause Support Enhanced' (niiiice) and 'Analytic Functions'. Under the 'Analytic Functions' they've got lead & lag. If they do what I think they do that alone would be a strong push to migrate up. A fair bit has changed under the hood and given that the damn thing is at core a DB engine, and all the sparkly is built upon that, it should have been mentioned here. Which leads me on to... MSSQL isn't up to scratch.
I don't call it mature, I call it decaying. When I joined my current company the first thing I was asked to do was to examine the feasibility of rewriting the underlying data management from MSSQL to mongo (yeah, really). They had some severe performance problems. I dug in and found it was the usual optimiser problem, failing on a ridiculously simple query something like -- select * from tbl where tbl.x = (select max(x) from tbl where...). Optimiser failed, caused massive table scanning carnage. It was so bad they nearly spent many many thousands to convert it to another (unproven) db engine.
Hardly the first headache we've had with it either. And the documentation is crap. Just for example, open BOL and go to <ms-help://MS.SQLCC.v10/MS.SQLSVR.v10.en/s10de_6tsql/html/a33d5db9-aed1-4c96-85bf-9d3b5051c635.htm> and try to find a function that returns the difference of 2 geographic instances. You can't cos it's not there (it's STDifference). There's a ***lot*** of bad shit in BOL, esp. on geographic data types, but all over.
TL;DR - MSSQL 2008 is beta quality and I have not expectation of 2012 being any better. There are also more worrying problems which I doubt even some of the writers of the article might know.
On to other points.
SSIS - my experience is more trouble than it's worth. If the data is bad then you'll spend almost all your time cleaning that and SSIS won't help. If it's good then you can load it in directly. Still, maybe my experience is jaded and it's improved, but it looks to GUI-heavy to make me comfortable.
Always on groups. Dunno. Power View. Ditto
DQS. Might be worth a poke.
MDS. Can't even understand what that's about. And "...with regards to IIS permissions and Silverlight"? I'm with eadon on this one. The smell of frustration, complexity, lockin and pain (pain is something MS is good at. Very good).
"What we are now looking at 14 years later with SQL Server 2012 is a mature set of tools" I don't bloody think so, and I suspect the various contributors either haven't seriously got their hands dirty yet or haven't got the guts to speak honestly about it.
I was too busy today to write my own diatribe, but the fact of the mater is that SQLServer is pretty piss poor at a lot of things.
SSIS (like DTS before it) is abysmal for anything beyond trivial, and even then it borks often.
To describe the optimizer as severely challenged, is as polite as I get. When you throw partitions into the mix, the optimization results are more often table scans than not. Add a few @vars in the predicate to spice up your life.
And don't start me on statistical drift ... brain dead statistics are a major pain and the optimizer relies on them and some hard coded values in the algorithms targeted at toy databases (which SQLServer still is).
I could go on about broken queries returning wrong data, broken internal structures never repaired, broken data type conversions etc etc.
Time for bed, enough fighting with this for today.
pjl
MDS. Can't even understand what that's about. And "...with regards to IIS permissions and Silverlight"? I'm with eadon on this one. The smell of frustration, complexity, lockin and pain (pain is something MS is good at. Very good).
The use of Silverlight and IIS in particular, with manual configuration required, makes me think it started as a skunkworks or internal project that someone then decided to throw into the product, without reworking the user experience. (To be fair, this sort of thing is very common in the data-analytics world, where lots of people routinely mess around with early-release OSS like the Julia language. I just spent a couple hours last night trying to fix a basic installation issue in Julia packages.)
Sometimes this sort of thing - a quick & dirty tool that gets released on the "make it work and you can have it" basis - can be quite good. Microsoft's kernrate sampling profiler and krView visualizer for it are a good example; kernrate isn't DTrace (what is?) but it can be very useful, and krView also comes in handy. But kernrate itself is a command-line tool with a boatload of obscure options and krView is an Excel spreadsheet. There's little user-friendliness there. (Of course, that's how some of us prefer to work anyway.)
I'm not saying IIS+Silverlight was the best plan for the MDS UI engine, but it wouldn't surprise me if that's the story behind it. Someone said "hey, I've been throwing this tool together, but it needs a UI", and someone else said "hey, I've been playing with Silverlight", and the next thing you know you have Reese's Peanut Butter Cups.
Whilst a little extreme, I do get your point. SQL Server is well known for having a really nice suite of management tools, but I was also hoping for some details on the database itself - at least as far as how the latest version compares to, say, Oracle, DB2 or PostGres.
Anybody got any pointers? I've not touched the last two versions of SQL Server at all, but have just been told that I may have to do some 'data quality' and 'migration' work as part of DB being adopted by our organisation.
I have just this week installed Win2012 and SQL2012 on a test machine.
I haven't had time to test anything yet though.
I doubt anything useful will have been added, and I doubt the various known broken things that are known as T-SQL will have been fixed. Nor will there likely be useful things, like an optimizer that understands partitioning correctly.
philip
The Redmond software giant understood that the simple storage of transactions was a given and that the era of analysis was about to start…
If only that (storing transactions is simple) were true! And also "dive through the SQL … provide an abstraction" SQL is an abstraction, though there is nothing forcing software clients to use it.
This article purports to be about a database but is actually about the subset of database work called Business Intelligence.
If your coding an application to support multiple RDBMS then many of the 'features' discussed are superfluous harder to support and more worryingly may tie your client to the platform and its upgrade cycle.
Nice I guess if your already in the club but this review doesn't make me want to join.
Microsoft is aggressively pushing it's Azure Cloud Computing with Hyper-V virtualization, all supported on Windows Server 2012 that has not proven, under any circumstances to be up to the challenge of competing with the likes of RedHat Linux or even FreeBSD OS that are preferred in supporting these leading technologies.
Just recently Netflix chose FreeBSD over Windows under every test for reliability, superior networking performance and scalability, and particularly rock solid security capability to power Networking Applicances for streaming millions of movies to clients each month. Every financial Stock Exchange in USA and internationally, particularly London Stock Exchange had thoroughly tested and found Windows Server, including 2012 to be inferior to RedHat linux for the significantly important trading functions that oversee billions of trades each hour involving hundreds of billions of dollars. They obviously could not entrust such critical technology base to a Microsoft product.
Windows Server 2012 maybe an improvement over Server 2008, but is not match for the incredible *NIX operating systems (OS) software that rightly dominate the server landscape today.
The docs do actually document the API for programatically controlling fuzzy match operations in DQS - it exposes a web service that's used in MDS integration. However, they also go to some trouble to emphasize that these APIs are not supported for application code.
Why might this be the case?
If you look at the feature comparison docs for SQL Server, you will find fuzzy matching components that do something very similar to the fuzzy matching functions in DQS. Catch is, these features are only available on Enterprise Edition. DQS is available with the (much, much cheaper) business intelligence edition.
I wonder just how late in the game their marketing wonks twigged on?
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
