March 2008 - Posts
I was following a debate on the new Merge syntax in SQL Server 2008 with interest. It was somewhat of a debate about hot things should work, what happens with NULLs, etc. Now I don't want to recreate it here, but one thing was interesting to me. What would you expect from this?
Merge (select top 1 * from xxx where yyy)
when matched ....
when not matched ...
..
How many rows should be inserted? Would the top 1 that matched each section be inserted? I'd think only 1 row would be processed. To me the MERGE statement should only have 1 row input to it and whichever section it matched would process it.
Is this not intuitive to anyone else?
I went down to the Denver stop on the Heros Happen Here launch events and as you can tell by this post, I left early. The IT Pro section was in the am and the developer section in the afternoon, but I grew a little tired of the developer keynote and after watching the SQL Server sessions in the morning, I decided I was better off leaving and not fighting traffic than sticking around to learn more.
I have to say I was a little disappointed with the morning session. I did get to see some people and spend time talking with them about various issues, and that's always something I enjoy. I learn some things, hear what people are doing and more importantly what they think about SQL Server. I get ideas for editorials, content on the site, and even how to better focus things.
The event, however, was really a marketing event, with canned demos and lots of slides that look to highlight the cool features and Wow you into buying or upgrading. Not necessarily a bad thing, but there was a shortage of good information given. Especially in the SQL Server sessions.
Kevin Cox and David Gollub tag-teamed the SQL sessions and these guys are both good presenters. However their demos were blowing up, less than half of them working, and that was disappointing. It tended to make SQL Server 2008 look like it's not ready for prime time. It's not even feature complete right now, but it looked really bad today. No major errors, but the demos didn't work and with two smart guys up there, it makes me think that there is really too much complexity to some of the new features. Policy Management, which I've barely scratched the surface on, was very confusing and tripped these two up a few times. Then the CDC didnt' work, which I've had trouble with and scratched my head on more than a few times. There was also the data mining add-in that disappeared; not something that you want to have happen when you're deploying or even demo'ing in your company.
I'm sure these demos worked in practice, but them not working live shows that there's a lot of work to be done.
On the flip side, the Windows Server Core demos were neat and I liked the idea of virtualizing a number of Cores that handle specific tasks, each separted from the other.
I'm starting to see results from other people testing compression and it seems that the data compression can really provide some nice benefits. I did minimal testing since I don't have any great, large databases and I'd have to create a load, which I didn't want to do. However some other MVPs have reported some nice time savings in queries, especially with COUNT().
However this is an Enterprise only feature, and I think part of the reason, along with the Resource Governer and Policy Management being limited to that edition, is that MS wants to try and get people to buy Enterprise. It's a huge revenue increase for them. My guess is that they see this as a way to increase revenue by converting people on the fence between the edtiions. After all, if they put everything in Standard (feature-wise), then why buy Enterprise?
However I think on a few things they're missing the point. If you added these features to standard, then you are a much, much more attractive platform than Oracle or DB2 at that low price point. Dont' forget that many people have multiple servers and the licensing costs can add up. My vote is that you move "down" a couple features every version from Enterprise to Standard. I haven't seen anything move this time (memory limits went away in 2005 Standard), but I'm hopeful something will change.
Oneof my tasks for our mini-Summit for Red Gate was to show the crew how I do the podcasts. There were fiveof us there and the hope is that at some point 2 or 3 of us can do video, so since I had the most experience, I brought along the camera, microphone, etc. If you're interseted, the final product is in the Monday's Database Daily podcast.
It was an interesting experience. I got everyone to work on a paragraph of content, since this newsletter is mainly commentary on the news. We struggled a little to get that done. I shot my piece first, showing them how I do it, Tony went next and did ok. Andrew did a piece, but didn't want to be in the final product, though I managed to snare a blooper from there. Anna refused to participate and Brad was nervous, but protected by his non-compete for a few more months.
After editing, which was fun looking through for mistakes, I showed them the process and how complicated it can be. I don't think it's that bad, but as I explained it, we realized there are a lot of steps and reducing those steps would be a great goal.
I'm not a bg fan of long meetings. I used to try and plan an hour a week staff meeting for my group, and then desparately try to cover everything in 30 minutes, take some questions, and then get everyone out of there. I hated being in 2-3 hour meetings as a manager, especially reguarly and I used to have 6-10 of those a week.
So with people from Red Gate in town this week, I wasn't sure that we needed today (Tues), Thur, and Fri as dedicated meeting days. However we've never all met face to face in one room since we've officially been the "Publishing Department", so this was a good meeting. Everyone wanted this in Cambridge, but I'm kind of a pain on travel, so we ended up in Denver.
It was a good meeting and we had some good debates/arguments about what directions to take with our 4 sites (SQLServerCentral, Simple Talk, P-Invoke, and Database Weekly). We also have a couple other sites we're talking about as well, and whether to merge them into an existing site or make them separate.
We didn't agree or come to any conclusions, but we certainly had a lot to think about. I think this was a good use of money, though I wasn't sure last week.
SQL Server 2008 brings with is a new set of features involving compression. At first I was assuming this was only backup compression, but as I've been digging in, there is a lot more to it. In fact, there are actually quite a few types of compression. We have:
- Backup Compression
- Row Compression
- Page Compression
- Log Compression
The middle two are actually under the heading of Data Compression, and what's even cooler is they follow the pages and rows to memory, so it's a boost there if your data is actually compressible. I haven't done much with log compression, but it's designed to make the transaction logs with mirroring go across the wire quicker.
On Wed the 5th, the Colorado PASS Code Camp takes place and Igot corralled into speaking. I picked compression, so the last couple weeks I've been reading through Books Online and testing a few things to see how well it works. I've also been getting distracted, so I spent much of this morning and today trying to get the presentation finished. I think I've got it, now I just hope it goes over well.
A couple interesting notes: first, these features are only in the Developer and Enterprise editions, so I'm not sure how widely they'll be deployed, but I bet they really help those large systems. Second, they're simple to use. Not a lot of settings, not a lot of experimentation to do, just enable or disable them. The data compression options are by table, index, or partition, but not necessarily all of them, so you need to enable it specifically for all areas. I bet people miss out on that a bit.
The last thing that caught my eye is that if you replicate a compressed table, the distribution agent doesn't check for a downlevel client, meaning a 2000 or 2005 client. So the table creation fails. How bad is that?