Too Much Data? Then 'Good Enough' Is Good Enough

ChelleChelle writes "While classic systems could offer crisp answers due to the relatively small amount of data they contained, today's systems hold humongous amounts of data content — thus, the data quality and meaning is often fuzzy. In this article, Microsoft's Pat Helland examines the ways in which today's answers differ from what we used to expect, before moving on to state the criteria for a new theory and taxonomy of data."

Comments

Obligatory

[Lifyre]

Obviously 640k was "Good Enough"

Seriously though he makes a good point. If you have so much information that it isn't stored consistently, with varying standards, or is an open field to be populated by an individuals perceptions. The example of all of the different colors of green (Green, emerald, asparagus, chartreuse, olive, pear, shamrock) is a great example of how one piece of information can be expressed in multiple ways. While you can define the color by using the hex code for it that isn't exactly an elegant or user friendly method of input or output.

He talks about various ways to handle these types of information from limiting input options to finding patterns and using those to "correct" the data.

Re:Obligatory

[Fluffeh]

It's not that there is too much data. That's not a problem at all. From my own experience (I work as a senior analyst for a multinational retailer employing around 200,000 people) it is rather that there isn't a single plan to utilize all the data we have available. Every time we introduce a new system or change the way we do something, the project inevitably drops a new table into our data warehouse. Now, this may seem like an acceptable way to do things, but after this has happened twenty times, it is nigh impossible to run a query that will return data from all these tables in any sort of reasonable time.

Would it cost more time, effort and money to properly introduce the new data to proper fact tables each time? Of course. However, the benefits would be that we could stop pretending that "we have too much data these days..." - because we don't. We just have too much mess with our data and it becomes unusable.

In the example above (different descriptions for green) the base system may need these particular terms, but if the data needs to be aggregated or used in another system, then the jobs that pass this to your data repository need to make those changes to adapt the data to work with the rest of your data warehouse. Having said that, if the new system is being developed inhouse, then during development the question should be asked "Can we store the color information in RGB right off the bat and adapt our own system to mask these values behind pretty descriptions?" rather than having to later do it via an ETL.

Re:Obligatory

[icebike]

It's not that there is too much data. That's not a problem at all.

Often, (more often then not, I contend), there is indeed just too much data.

Because we have all these marvelous computerized data capture system doesn't mean the data is necessary, useful, or worth keeping. However, someone always comes along in the project design stage and insists the millisecond by millisecond weight of a bag of popcorn weighed in real time as it is being filled is going to provide a wealth of data for the design of future bagging systems and materials handling in general.

The scale was only there to assure that 10 pounds were in the sack and to shut the hopper. Then some fool found out it measured ever few milliseconds and recorded the data.

So the project manager gets brow beaten into recording this trash which invariably never gets used for anyone for any purpose at any time, as those who lobbied for it wander off to sabotage other projects and never revisit the cesspool they created.

This happens way way more than you might imagine in the real world these days.

It used to be projects had to fight for every byte of data collected, there were useful sinks identified for every field. But with falling storage costs the tendency is to simply keep shoveling it in because its easier than dealing with the demands by those "researchers" looking for another horse to ride.

Re:Obligatory

[StuartHankins]

+1 Insightful. I would argue that -- just like you have a lifecycle for software development -- you have a lifecycle for nontrivial amounts of data. Some data is useful in detail for a short term, but wherever possible it should be more coarsely aggregated as time progresses, and you should get sign-in from executives that it can be dumped after a period of time.

Where I work, I estimated the cost to upgrade our SAN to continue to store a set of large tables which helped everyone understand the cost in real terms. People tend to think once the data is imported or created that it's a small incremental cost to house it from that point forward, but backup times and storage along with execution plan costs increase with size. There is a performance benefit to this trimming; partitioning and check constraints will only get you so far.

What is difficult to gauge in advance sometimes is how the data will be used -- some things are obvious in the short-term, but as the company looks to different metrics or to shine some light on an aberration, you really need to be able to determine how quickly you can dump the detail. Get signoff then add some padding so you are conservative when you destroy. Make a backup "just in case" and delete it after a few months. The good news in my work is that changing your mind later to adapt to the new requirements means expectations are already set to change the way it works "from this point forward". There are many fields of work that do not have that luxury, because of the time or cost to gather detail again.

Re:Obligatory

[smallfries]

It's not a new idea, it's been explored before and it only works in certain cases. Take a look at Ontologies are overrated. From the section called "Mind Reading":

You can't do it. You can't collapse these categorizations without some signal loss. The problem is, because the cataloguers assume their classification should have force on the world, they underestimate the difficulty of understanding what users are thinking, and they overestimate the amount to which users will agree, either with one another or with the catalogers, about the best way to categorize. They also underestimate the loss from erasing difference of expression, and they overestimate loss from the lack of a thesaurus.

Here's the one line summary of TFA:

[billrp]

SQL DBs are not appropriate for storing, processing, querying, and browsing unstructured documents.

Re:Not just large sets of data

[QRDeNameland]

From a quick skim, it seemed to be yet another "SQL/RDBMS is dying because we have too much heterogeneous data to handle", and a rather ambling and long-winded one at that.

And there was much rejoicing... "yay."

[VortexCortex]

A bunch of rambling self-evident or speculative statements, followed by conclusion:

Conclusion

NoSQL systems are emerging because the world of data is changing. The size and heterogeneity of data means that the old guarantees simply cannot be met. Fortunately, we are learning how to meet the needs of business in ways outside of the old and classic database.

Which was apparent to everyone, and missed the real point: We have lots of data, and we're too impatient to wait for it to be aggregated, synchronized and processed. There goes 10 minutes of my life I'll never get back.

Here's a hint: People working on the solutions to this problem work in the financial sector and in quantum physics.

Too Long; Do not Read

[Comrade Ogilvy]

The researcher is just throwing together a bunch of problems that have existed, in some fashion, for a very long time, and concludes with open questions rather than even vague proposals for solutions. So I would say this article is both too detailed, and not detailed enough.

Re:GOATSE ALERT

[Anonymous Coward]

I do,
I have for example the http://www.thoughts.com/geekatwork/at-work.
I am putting goatse onto various wikis, blogs, etc. Works like charm.
Lately I somwhat busy so I use that tinyurl link
(but that link isn't that simple, it contains 'data:' blob which decodes to tiny html page that embeds goatse.
So that should fool extensios that resolve shortener links.
Anyway, todays trolling got me 279 victims. I can do better that that sometimes.
(In total I got around 10000 victims on goatse.ru so far, and handful of hundreds on tubgirl, some random gay porn link, hai2you, but I prefer to post goatse.

I also keep a collection of responses (both positive and negative) - aka troll food.

Favorites:
"Ugh. Goatse. NSFW. Asshole (poster and picture, both)."
"Seriously ... new account to post that ... what a douche!"
"You're a fucking douchbag." - "That is the most accurate comment yet"
"I hope you die in a fire before you are old enough to contaminate the gene pool."
"Death to all assholes - Let's put you first into the guillotine"
"Asshole... Ginormous asshole, in fact."
"Ugh. Goatse. You asshole."
"Better than you, you arse bandit."

Hate:
"I hate your guts."
"WTF you fucking asshole."
"Damn! Mod this fucker to hell"
"Fucking troll, do not click there"
"It would be more interesting if I had a piece of pipe and your face, in close proximity so I could smash your face beyond recognition,"
"You fucker" - "I had the same thought as you. What a fucking asshole. The link is nsfw."
"Bravo teeny bopper. You're a really mature mother fucker (or do you prefer father fucking? Damn you homo erotic shittter)."
"Wait! I think I hear your mommy calling to give your tongue a good soap washing. And maybe she'll execute you too"
"You fucking piece of shit!" , "You sorry piece of shit." , "You cunt.", "Fuck you."
"It's because of Assholes like you that I can no longer trust URL shorteners"
"I did not even bother to look, but this same idiot has been doing this for weeks now. Fuck off asshole."
"What a retard..... enough said...."

Funny:
"Didn't click it, but the magic 8-ball says goatse."
"Thanks, I'm reading slashdot in class like a good student and just got tubgirl'd."
"not gonna click it to find out, but I'd be surprised if parent's link wasn't goatse... It appears you would be correct sir. Why oh why do I always forget.."
"Watching second monitor, there was something wrong with the other screen. Control + w. Phew..."
"Doh! One has to also recognize data urls. *sigh*"
"That's somewhat clever, but some of us do know what base-64 encoding is."
"Can you not afford normal entertainment?"
"Hey family! Come look! They're opening the Google Talk client! Now, click here...... (sees goatse)"
"I tried to post warnings about the goaste loving jerk yesterday but was modded into oblivion as a karma whore"
"Turn on TinyUrl previews. It saves lives."
"Posting your picture online again?"
"Really? Are you not tired of this yet?"
"High likelyhood of being a Goatse link. Proceed with caution"

Emotion:
"i WAS eating lunch you ass!"
"Oh dear god my eyes. Haven't seen THAT awful image in a while."
"My eyes are burning... argh! Damn you!"
"MY EYES... dude i am at work here "S "
"WARNING: Don't click on the parent's link! Damn goatse! The first I experienced, no less.
"Oh goddammit. I didn't need that right before bed."
"goatse warning! I'm still recovering."
"Please friends, I beg of you, do not click that link! Do not look at that image, whatever you do! It is a bad image! It is a goatse image."

Frustration:
"Can someone make a fucking goatse blocker firefox plugin please? This is pissing me off now."
"I am sick and tired of that crap on /. "

Philosophy:
"Goatse trolls are getting better these days..."
"Why the sudden coordinated campaign for Goatse? Is someone making money off this?"
"You're right, this is the most

Confused and incomplete

[lucm]

This article is confusing because most of the verbiage is made up by the author (such as "inside" or "locked" data). It is also misleading because it seems to indicate that structured and unstructured data usage is the same. Well it's not - a very large proportion of unstructured data is blog posts and emails but the amount of search and aggregation that is performed on this type of information outside of a few major companies (such as Google) is very low, which makes this usage a niche and not a trend maker.

The reality is that there are three categories of data that are relevant for databases: numbers, text and spatial. Everything else, which falls under the umbrella of "binary", is very unlikely to benefit from a database engine; only the metada can be manipulated and this metadata falls under one of the other categories and is a very good target for ETL. And so far nobody came up with a reliable way to search binary, such as video or audio, without relying on heavy indexing, metadata or any kind of transformation that takes binary and make it text data.

If a piece of data cannot be searched or aggregated, it does not belong in a database, it belongs on a filesystem. Anything can be done with blob columns but performance is usually not very good because the database engine cache is not designed for large objects. NoSql or not.

Also there is so much happening with storage infrastructure, such as sub-volume tiering or block-level replication, any analysis of data that does not take a look at storage is flawed.

Any slashdotter coulda told him that.

[Anonymous Coward]

We don't read articles, just skim the headline, maybe the submittal, and then a few top ranked posts.

That's Good Enough! (tm)

Re:Illiterate title

[cforciea]

Either usage is acceptable.

Statistics with Comp Sci a kick-ass combo.

[PerlPunk]

This is why Statistics will become more and more important over time--it allows you to make inferences about populations that you couldn't possibly count. If you already know Comp Sci or or learned how to program on your own, go for a couple of Stats degrees. Along with your programming skills Stats will do you very well as the information age unfolds.

Re:Not just large sets of data

[interkin3tic]

I skimmed the article, and I can say this much: they mean a fairly specific type and use of data.

Too much data from scientific results? Only the researcher him or herself would ever say there's "too much data." Everyone else says "not enough data." Everyone. At all times. Especially his committee and reviewers. Even when I've worked so hard for so long for so little money. After all, THEY'RE not the ones who are sacrificing their happiness, time, effort, hairline, and relationships to...

Uh, I mean, yeah, the article is vaguely worded and doesn't apply universally, and my life sucks.

Nothing new

[Whuffo]

If the people that write these stories would familiarize themselves with Information Theory (Claude Shannon, in the 1940's) then they'd understand that you still can't make silk purses from sow's ears.

Yes, it's a lot of records. Yes, the data entry people made mistakes. All this really means is that there's more noise in the data. As the signal to noise ratio declines, the value of the results also declines. Making decisions based on noisy data isn't science, it's only guesswork. That's fine for weather forecasting (a similar problem) but expecting the results from the described data to be more accurate than weather forecasts is foolish. Remember: garbage in, garbage out.

Ambiguity Management

[AtomicSnarl]

The problem being encountered is one I've faced often in 30 years of weather forecasting: Ambiguity Management.

The weather business deals with reams of data from thousands of sources and all the complexity of trying to follow a single swirl within a flowing river to figure out where it will be tomorrow. Decades of research and modeling have evolved into dozens of primary rule-based tools available to forecasters which are applicable to most situations. Objectively, you should be able to follow the rules, weed out the conflicting or contradictory ones, and get a reliable answer. Realistically, you don't. Why? Two reasons:

1. The dataset is incomplete.
2. The tools are imperfect.

You simply can't have perfect knowledge of all the relevant details in the atmosphere to feed a completely objective tool (computerized model or whatever) to get your perfect prediction. Like Rosanne Rosannadana's mother said, "It's Always Something!"

The trick then in being a good (aka reliable) weather forecaster then is how you manage the ambiguity of incomplete data filtered through inherently biased tools. Some weather stations run hot or cold, have local effects enhancing or reducing pressure or winds, etc, etc, etc. Good models account for this, but that's a static adjustment, not a dynamic one. Models run hot or cold, fast or slow, depending on their structure and assumptions, and they reval their strengths and weakness over time compared to other models and reality at verification time.

The basic forecasting questions are - Where is it, Where is it going, an what will happen when it gets there? Because the models are perfect (100% replication of output from identical starting states), but are always wrong (inherent model and data limitations), you make your money examining the consistency. The model(s) are running slow and cold recently due to the whatever event going on? Ok -- warm it up a few degrees and expecting things a few hours earlier than it forecasts tomorrow. Some models handle well in winter but get klutzy with large thunderstorm events. One model I worked with covered the world in clouds if you waited long enough. Solution? Don't trust it past X number of hours. And so on for the family of models through the decades and to today. Some models have high skill up to a certain point then it drops off quickly. Others show less skill, but are decent for the long haul. You get the idea. You can make a forecast using only one tool, but you can make a better one using several and sorting out their differences by using ambiguity management.

Needless to say, you needed a solid understanding of the physics and dynamics of the atmosphere to help make good decisions to do all this effectively. The modelers and users now data mining these huge collections of information likewise need a solid understanding of Statistics and the event mechanics they're examining to make any good sense of it all. At the very minimum, a large poster announcing "Coincidence is not Causation" needs to be in every office, otherwise you start getting breathless announcements about how underarm deodorant "causes" cancer because people eating hamburgers had a lower incidence rate by comparison.

Your Mileage May Vary -- a lot. That's the point.