PASS Summit 2009 Key Note 3

November 5, 2009 at 1:24 pm (PASS, SQLServerPedia Syndication) (, )

Dr. David DeWitt

“I’m not a doctor.” This is going to be good. “From 1 to 1000 MIPS” He’s doing great. He was a fantastic last year and I knew this was going to be good. 32 years in academia and only in MS for 1.5 years. He’s the blue sky guy, and beleive me when I tell you, you get smarter from being in the room with him. And a huge ovation because he told us that he’s going technical and not covering a marketing pitch.

Right. This one is going to be hard to blog. He’s going through information quick. It’s all good. He’s giving an academic talk about the 30 years of technology trends in databases. He’s going over how the trends have affected OLTP and why the trends are forcing DBMS to evolve and some future trends to keep an eye on.

Back in 1980 we had a Digital VAX with a 1MIP CPU, 8mb of memory, 80mb disk drive… all for $250K. Yeah, the “good old days” Not. More or less 1000 times better for Cache & memory, 2000 times better for CPU and 10000 times for disks. But transfer rates have only increased by 65X. And for seeks there has only been a 10x improvement. 1985, 100TPS & 400 I/O/second. Today, 25,000 TPS & 100,000 I/O/second. His key point is that it took 14 drives to keep a single processor busy in 1985 but it takes 300 drives to keep a cpu busy today. He believes that SSD and phase change memory are the only things that are going to make a difference here.

His comparison of the problem is that today we’re in the situation where we have a gigantic water tank to feed water to the hole town, but only a garden hose to drain it. That’s scary to think about. When thinking about it, he compared the disk to cpu ratio and came to the conclusion that drives are 150x SLOWER today than they were 30 years ago.

So his basic conclusion is that we need to avoid random I/O in our database applications…. uh, yeah, good advice, I guess, but how the heck do we do that? Obviously we don’t and that’s not what he’s saying. He’s making a great point. This is just flat out fantastic.

In the old days it was 6 cycles to access memory. Now it’s 200 cycles to access memory. Despite all these improvements, we’re really messing up our scalability on the hardware because of limitations in the design. Hard to believe when you consider that everything is, in fact, faster, but it’s just not faster the right way. It’s scaling too hard.

He’s showing how the queries work within the system, what happens, literally, what is waiting on what, all the way down at the lowest level of the process. Scan queries vs. join queries. So why so many stalls? It’s the L2 data cache stalls, an artifact of the last 30 years of design within systems. Why? Well, you’d need to be here to really understand. I’m not going to repeat it correctly. For each record, there’s an L2 & L1 cache miss from each page as it reads from the disk, row-by-flipping-row. So basically, we’re getting RBAR within the CPU when it has to read from the disk.

There is a new memory configuration called a column store where the columns are broken down instead of rows. The columns are stored on files/pages. He’s showing how ID would be one page, Name on 2, City on 12, etc. because otherwise the rows are stored across pages. Without compression, it changes the amount of data on a page. If they also add compression. It’s going to make a huge difference. BUT, it’s not updateable. They’re going to make the DB a perfect decision support engine. Showing some drill down on the details, he showed about a 7x improvement in speed, without compression.

Then, you put compression on top of it and stuff really takes off. But remember, this is a read win. It’s write lose.

Really, I’m going to stop trying to keep up. I’m starting to lose track of everything anyway. I think this might be available online or on the DVD. I sure hope it is. If you were at the Summit and you didn’t attend this… you messed up, big time.

The key to compression is to remember that you need to get more data onto the disk because the CPU is 1000x faster but the disk is only 65x faster. That’s why you compress. Not because you’re saving space, but because you’re saving READS. I’m going back to the office to turn on compression on everything (after testing & verification).

I can’t describe it all, but the early/late materialization performance differences are amazing.

Key points for the quiz. yes, I’m giving out the quiz answers, call Bluto and D-Day:

  • At first glance hardware folks would appear to be our friends
  • Huge inexpensive disks have enabled us to cod-effectively store vast quantities of data
  • On the other hand ONLY a 10x improvement in random disk access and 65x improvement in through point
  • Two pronged solution for “read” intesnsive data waroehouse workloads – parallel db technology & column stores
  • Column stores, minimize transfer, facilitate compression, minimize memory stalls
  • But column stores don’t work for OLTP, AT ALL
  • Hardware trends and demands are forcing DB systems to evolve through specialization

But, I missed to two bullets because I couldn’t type fast enough. So you might still fail the quiz.

This will be on the DVD. Thank you PASS. Thank you Dr. DeWitt.

Permalink 1 Comment

PASS Summit 2009 Key Note 2

November 5, 2009 at 12:10 pm (Uncategorized)

Patrick Ortiz from Dell, the platinum sponsor for the PASS Summit.

The goal will be to discuss the Dell Microsoft Practice, Configuration Management, Disaster Recovery and Consolidation. The MS Practice team has server tools, messaging & communications and collaboration & databases. Lots of work going on there.The see the CM, DR & Consolodation as linking cogs, if you trust the slide. They want to start implementing CM management systems. He’s driving the data suggested with Asset Data and physical server information. Then you drill down and collect SQL Server information followed by drilling down further to collect DB info. He’s gone on to show some of the benefits of CM. The idea behind CM is so that, once you’ve defined your servers, you can define disaster recovery and start looking into server consolidation. I hate to say this, but this is really the equivalent of eating your vegetables. This is pretty dull stuff. A key note needs to be a little more exciting than this. Show us some hardware, show us some flashing lights, some screens, something. It’s early in the morning. More.

Permalink Leave a Comment

PASS Summit 2009 Key Note 1

November 5, 2009 at 11:50 am (PASS, SQLServerPedia Syndication) (, , , )

Wow, the bloggers table is empty today!

Nice intro! Good photo’s. I love the Summit!

Bill Graziano is introducing Day 3. He acknowledged Twitter and the bloggers. We rock! Outgoing board members are Greg Low and Pat Wright. These are great guys who’ve busted their butts for the community. Kevin Kline is completely off the board now, finishing his time as the immediate past-president. I’m pretty sure that’s the first time he won’t be on the board. Yep, I’m right, he’s never been off the board since PASS was a organization. He really has done a lot for the organization. Thanks for your time Kevin. A review of all the other board members including the new president Rushabh Mehta and Wayne Snyder as the immediate past president.

April 21-23,2010 the European Conference will be in Neuss Germany. Next years conference will be in Seattle (again…. blech!… I understand why, I do, but it’s a ROYAL pain to fly across the whole continent every year). A thanks to the headquarters staff too. These guys are great and they work some odd hours. You’d think they were DBA’s. Nice work guys.

Permalink Leave a Comment

PASS Summit 2009 – Day 2

November 5, 2009 at 8:53 am (PASS, SQLServerPedia Syndication) (, , )

Day 2 of the Summit was extremely busy. I missed a breakfast on DMV’s that evidently was one of the hits of the show. Day 2 was kilt day. I wore mine and there were two others, Steve Jones and Bill Fellows. They were a hit. Next year I’d like to see more. There are pictures all over the place. Track one down on your own.

I set up at the bloggers table and blogged & tweeted my way through the key note. Except for the hyper-sexy Windows 7 touch screen computer, it wasn’t the most exciting key note I’ve ever seen. The new technologies coming out for data manipulation on the client machines in Office 2010 are impressive, no doubt, but something seemed lacking and I’m not sure what it was. During the key note they gave out awards to the outstanding volunteers of the year and they gave the PASSion award to best of the volunteers. I won one of the outstanding volunteer awards for the work I did with the Editorial Committee & SQL Server Standard. Allen Kinsel was the winner of the PASSion award. I wanted Allen to win. I had put in a nomination for him and I knew others that did as well. Congrats Allen. You earned it.

My first session was completely personal for me, and I doubt most people would be as excited about it as I was, but I was very excited to see Michael Rys of Microsoft talk about spatial indexes. It was a very useful session. I took a lot of notes, but I’m going to have to go back and watch it again on the DVD because there was stuff I missed. If you are working with spatial data, you should check this one out.

Lunch today was just lunch, but during it, many of the MVP’s that took part in the MVP book project, SQL Server MVP Deep Dives, all proceeds of which go to War Child International, were signing their chapters in the book. I bought the book, but only went over afterwards to get it signed by the last four or five that were still there. I told Paul Randal how disappointed I was that he didn’t wear a kilt. But, during his session he did put up a picture of himself in a kilt. He doesn’t own one. I hit on MVP’s for the rest of the day to get their signature. I’ll be carrying it with me today (and breaking my back, it’s huge) to get some more. Buy the book. You’ll get good information and help an important charity.

After lunch I went to Ben Nevarez’s session on How the Query Optimizer Works. It was very good. I took some notes and got some ideas for future research. I really like the way he did simple, direct examples that illustrated his point. He also reminded me of a DMV that I should have included in my session (and will next week). This is useful stuff and it’s definitely worth checking out on the DVD.

I skipped the next session to go and do a video interview with the Midnight DBA. For those that don’t know, the Midnight DBA is actually two people Jenn and Sean McCown. It was a blast. We talked about books, TSQL, Kenpo and sexual harassment of men by men. Yeah, it was pretty fun and not entirely appropriate (although, probably only PG in rating and completely work safe).

After that I did a thing with Microsoft, completely unrelated to being an MVP, but still NDA. Done.

To finish out my time at the Summit proper I went to Buck Woody’s session on using Policy Based Management, Data Collector and Central Management Servers. I’d met Buck in the hall and he asked me to come down. I’ve never seen Buck Woody speak before. For those who know, why didn’t anyone tell me? All I can say is, what have I been doing with my time at these conferences? I should have been going to see everything the man says. Not only was it useful information, but it was hilarious. We were laughing our butts off through the whole thing. He picked on people in the audience (especially poor Aaron Bertrand, although I took some hits when I showed up late, wearing a kilt), cracked jokes, and gave out useful information in a clear, engaging manner. I’d suggest you watch the DVD, but you’re not going to get the full kick in the britches that being there will give you. However, you will still see some good stuff. I’ve already sent off a message to my team lead to suggest we get started with some of this right away.

After the show was over it was party time. I had two to go to, but I only made one, the Microsoft hosted party at GameWorks. Good food, free alcohol, video games. I rode the Hummer with Wendy, @wendy_dancer, driving. She drove better than I would have. It was fun, but I went home early because I’d had a nasty cough and runny nose all day and my voice was completely shot. I needed rest.

Another great day at the PASS Summit 2009.

EDIT: Updated Bill Fellows name.

Permalink Leave a Comment