Parallel SQL Server?
-
I caught this note that talks about the next version of Windows being fundamentally different. I thought this might occur years ago when it seemed there was more work being done on clusters of parallel processors, but that never really took off in the mainstream.
But now with dual and quad cores becoming commonplace, and the chance we'll see 8 or 16 cores in the next few years, this seems like it might be a good time for Windows to start taking advantage of multiple processors on the desktop. Especially if it can get Word, Excel, IE, and other mainstream apps to run quicker from the user perspective.
But it also makes me think that more and more of the SQL Server codebase will be written to run very efficiently, expecting multiple processors to be visible. The -P startup switch may truly be a thing of the past after SQL Server 2008 or the next version.
And maybe then we'll start to see some sort of parallel cluster with SQL Server running on 2, 4, or any number of nodes. After all, there are some others working on this problem already: PCTI and ParAccel among others.
I also saw this week that the PDC is being cancelled for 2007. Does that mean what this article says? That Windows 2008 is being delayed? Who knows, but I think that the PDC is a big event and there are plenty of developers that have planned, gotten budgets, etc. Canceling the conference might make sense to coordinate with technology launches, but it's a hassle for us. There should still be plenty to show with Silverlight, Katmai, and more by then. The official note here.
But even if there's no PDC, there's still the All Things Digital conference with both Bill Gates and Steve Jobs on stage. What Would You Ask Bill and Steve if you could? The show is sold out, but it should be interesting.
I also got an announcement recently about the TPC-C and TPC-H Benchmark Results for SQL Server. Apparently SQL Server kicked some tail in the 3TB benchmark for warehousing. It's not the lowest Price/Perf number, which was surprising, though it is the lowest non-clustered system. The lowest is a Linux/DB2 system that's clustered. How a clustered system is a lower cost stuns me and why isn't a Linux/DB2 non-clustered system the best price/performance? I'd like someone to explain that.
Steve's Pick of the Week : Tony Rogerson, SQL Server MVP in the UK, comes up with some very interesting posts, but I thought this one was very, very informative. If you write T-SQL, you might want to check this out. -
Steve -- first, let me say that I follow your stuff, and what you do is great work. Let me explain what you want explained. First, it's all about divide and conquer. With multiple nodes owning their own data, the memory and I/O channels can be smaller, I can use more commodity components and really scale (linearly). I won't go into the technology, but it's very similar to Teradata and what the appliance vendors are doing. If MS wants to scale, they are going to need scale-out -- and I"m surprised they don't have it since the guy that literally wrote this technology for DB2 left for MS like 8 years ago. They are working on its however for post Katmai.
As for price/performance, they were saying at the MS BI conference this would be a leadership position - it isn't. I found it odd they would announce this and show this result.
One reason why MS is so expensive -- premium processors -- a 32-way Itanium is not a cheap box. But let me burst the MS price/performance bubble even more. AS well, it just can't produce the throughout of a shared nothing scale-out (don't confuse this with RAC -- if you compare those marks, they get poor scalability).
The DB2 result includes upgrade protection and 24x7 support. MS includes none of that. So you'd have to add 25% to the software price and then about 11% for hte 24x7 support depending on the MS support package. MS used to price in 3, 12 a year call backs at about $1950 a year -- they pulled a fast one on us a couple of years ago and moved to a singel call of support in 3 years!!! Not realy world for me.
Also note how this is licensed. The DB2 result is licensed for UNLIMITED users. SQL SErver -- looks like about 70....some difference. Now make SQL Server unllimited and what happens -- it gos through the roof!
As well they didn't even use range partitioning? Why -- well, any scans that affect less the sum of all (or a single) table partitions wouldn't be parallelized. They also turned off stats collection and some other stuffF
OK, I'm not trying to bash -- and I know these benchmarks very well. Oracle is the WORST of all when it comes to misleading and MS is a close second. That DB2 result, like all TPC-H, has trickery, but it's at least acceptable. It's really sad to see all of this to be honest....I'm quite disappointed in cheap marketing tricks that obfuscate reality and the orgs that allow this to continue. Want a hint when you look at Oracle results? They use term licensing -- in other words, you don't even own the product.
One credit to MS, on their NEC 1 TB TPC-H result they didn't even RAID the disk to keep costs down -- at least they did that here.
Anyway -- this is the reason why it is so cheap (the scale-out technology and commodity components). It's also a blatantly honest explanation on how the MS price/performance should be so much more if you want to compare apples to apples.
Did you see it took 25 hours to load that database? That's cause it doesn't scale-out. DB2's two years ago? 2 1/2 hours
BTW, I am underwhelmed by this result -- two years (a lifetime in the benchmark world, in TPC-C, leading result has gone from 1,000,000+ tpm to over 4.2 million!!!!!) it barely beat that DB2 results and couldn't beat it's price performance?
Llet me say that I work for IBM -- so if people want to take their shots at me, they are welcome to, I'd be happy to respond. And again, I didn't mean this as a bash -- it started out with they just need to work on scale-out -- I really like some things about SQL Server. But I think we will all agree, the pricing tricks are borderline obscene (1 call in 3 years when they used to at least price 36 calls in 3 years..and I won't even comment on the discount).
-
I dunno Paul. I have had to call MS excatly twice in 12 years for help with SQL.
-
I'm sure that there are some strange tweaks in the benchmarks. All of the submissions are geared towards promoting their own platforms.
I'll say this from my experience, which is limited on other platforms. Everything I've seen regarding pricing shows DB2 to cost 1.5-2x what SQL Server costs. Perhaps the client numbers change that and I'll admit that I'm not sure, but that's the raw pricing I've seen. Oracle tends to weigh in at 1.5-2x DB2, so I've always considered them very expensive. However they all have strange pricing that requires a PhD in Econ + some heavy calculas background to decode.
In the support arena, I've worked on SQL Server for 15 years and managed DBAs for Oracle/DB2 over a year. I've regularly called Microsoft for support, probably 10-20 times at different jobs, but I've also had hundreds of servers I've never called on. Overall they've had great support and I've had few problems that they couldn't resolve. Their support was on par with Cisco's, whose I consider to be the best of large vendors.
I've rarely called on Oracle, but when I have the support wasn't great.
On DB2, we had to call quite a bit on a few systems that were having issues. The support was good, but not great on followup and we didn't resolve the issues. However I'll say that I didn't manage enough systems to have this be anything other than anecdotal since we had issues on those few systems and I'm not convinced that it wasn't just the application.
Overall I'll say that it seemed that DB2 was harder to manage, requireing less servers/DBA and more work. I'd also say my DB2 DBAs had to have more skills than the SQL Server ones. It just required more thought. That was my impression. Not that SQL Server was "easy", just easier.
All this being said, I think that all three platforms are excellent. They will all do the job and can handle most loads. I think pricing and value are heavily dependent on your size and the negotiations that your company works through.
-
Hey Steve --- I agree with the pricing stuff, but the issue is that (a) no one pays the posted price, even TPC-H alots discounts for 40% on 25% and (b) it doesn't take into account the TXs per core. Case in point, SQL Server 64-way Itanium 2 TPC-C is like 1.2 million tpmC --- IBM beat that on a Power 6 with an 8-way box (16 cores). So I agree it's confusing. I've seen Oracle discount into the 90% marker, and MS too - -so the bottom line no one pays raw pricing and you need a PHD in Econ and some heavy calculus for sure (I have a BA in Econ, an MBA with lots of Calculus, but you really do need that PHD -- lol.)
I'm happy to hear that for support at MS -- in fact, I"ve heard clients often ask my why Oracle and DB2 include support without your choice (all except DB2 Express-C). I think MS has a very nice support model -- it's granular and lots of choice -- I'm just saying that to compare even prices you should ensure you are all buying the same thing.
I think you're impression on DB2 of old is bang on - I've personally drove development to address - outside the scope of this post - but DB2 v8.2 and especially V9 has some trumping stuff (eWeek gave it an Administrative rating of Excellent vs SQL Server 2005 of Good).
In the end - for most cases, all 3 can do well, and prices are right to. I do give all the credit to the MS BI proposition --- full kudos. I never post comments like this -- but for some reason felt compelled.
Anyway -- again, you have a great resource and I recommend it highly to my clients that have mixed environments (which is like 99% of them!)
-
What is the -P startup switch?
Live to Throw
Throw to Live
Will Summers
Viewing 6 posts - 1 through 6 (of 6 total)
You must be logged in to reply to this topic. Login to reply