May 21, 2008 at 11:54 pm
Isn't this basically what Oracle RAC is. You have many instances of Oracle running on individualy boxes, all accessing the single shared data source. They pass a token about to control locking, and each instance has it's own resources in terms of memory and cpu, and it scales in a similar way to a web farm. OK, I'm sure you can't add an Oracle instance quite as easily as web servers, but the principle is there. It's the MS "shared scalable databases" but Oracle allows all the instances read/write access to the data. OK, there is most likely a performance hit, I'm not an Oracle DBA so I don't know, but that is most likely a minor trade off to get the scalability that RAC offers if that is what you need. Is this not just Microsoft trying to catch up with Oracle on the horizontal scaling front, or am I missing a significant difference in what is being proposed with MatrixDB, that will make it very different from Oracle RAC.
Cheers
Roddy
May 22, 2008 at 1:45 am
Hmm. The thing with RAC is that however many nodes you have, you're still accessing one central database (shared everything, as the jargon goes). Microsoft's MatrixDB idea is supposed to use a "shared nothing" approach, so I suspect it's intended to work more closely to the model used by the old distributed Netware Directory Services database than anything else. That said, the mental image I have of how Microsoft'll do it is very hazy, since if I was clever enough to second guess them better, I'd be able to write a competing product 😉
Overall, I think it's a good idea, but the complexities involved mean I don't think I'd like placing any reliance on it in v1.0 form....
Semper in excretia, suus solum profundum variat
May 22, 2008 at 3:15 am
Components could fail, and it would seamlessly adapt without any downtime or any admin intervention. And no query or workload would be too big for it, because to get more horsepower, all you would need to do would be to add more hardware components.
These seem, to me at least, the two most significant sentances.
It would suggest that the future of db technology will not lie with a highly skilled DBA optimising code and managing resources to eek out every last ounce of speed from their servers. Rather that the mantra of chuck a chip at it or a few gigs of extra ram to solve the poorly running query will actually become the norm.
Lets face it MS are not going to cry the day they manage to design the self managing, self optimising and self healing database. I suspect that this is the first real step down that road.
K.
May 22, 2008 at 6:14 am
I think this is more of the Google model than RAC, trying to get some amount of copies of data for redundancy of access. It's an interesting idea and I'm waiting to hear more at some point.
Not sure they'll get to the self-everthing database, but they are trying.
May 22, 2008 at 6:41 am
The thing with RAC is that however many nodes you have, you're still accessing one central database
Maybe I'm missing something but if there is not one central database doesn't SQL Server already come close to this approach using replication to slave servers? The beauty of RAC is that you can put the database on high performance disks and easily and cheaply accommodate failure and performance needs by adding boxes to the RAC.
What I heard in this was some type alternative to Oracle RAC. I've been very surprised that SQL Server hasn't addressed the RAC. It's a nice solution with not much performance overhead and at our organization the ability to scale via the Oracle RAC vs buying bigger and bigger servers (we needed read/write real time data) meant that a project went Oracle that might have gone SQL Server otherwise. (We're a mixed environment)
May 22, 2008 at 6:47 am
They probably want something to compete with RAC, but be more accessible to the masses. They probably also don't really know exactly how to build it but probably have a few ideas, hence the job postings. I hope it works out and is included in the Enterprise Edition.
May 22, 2008 at 6:47 am
Samuel Clough (5/22/2008)
The thing with RAC is that however many nodes you have, you're still accessing one central database
Maybe I'm missing something but if there is not one central database doesn't SQL Server already come close to this approach using replication to slave servers? The beauty of RAC is that you can put the database on high performance disks and easily and cheaply accommodate failure and performance needs by adding boxes to the RAC.
What I heard in this was some type alternative to Oracle RAC. I've been very surprised that SQL Server hasn't addressed the RAC. It's a nice solution with not much performance overhead and at our organization the ability to scale via the Oracle RAC vs buying bigger and bigger servers (we needed read/write real time data) meant that a project went Oracle that might have gone SQL Server otherwise. (We're a mixed environment)
I don't say Oracle's RAC is a bad idea; just that it's based on a different paradigm. Eventually, it boils down to the difference between slower disc subsystems under less load vs higher performance disc subsystems under greater load. Which is best? Big fat "depends" is the right answer.
Semper in excretia, suus solum profundum variat
May 22, 2008 at 8:47 am
I think this is a RAC competitor, just like Scalable Shared Databases or replicating to slaves.
Log shipping existed in many installations before Microsoft built a support solution, this might end up being auto-replication of some sort.
I've heard RAC doesn't sell well, hard to administer, expensive, etc. Perhaps this is way to solve some of those issues as well.
May 22, 2008 at 9:22 am
The use of the word "Matrix" makes me wonder if MS is doing (or has acquired) some of the distributed lock and storage management techniques from Polyserve. This certainly has the potential to work like Oracle RAC, and lack of shared everything clustering has been a knock against MS in the past.
May 22, 2008 at 9:24 am
Steve Jones - Editor (5/22/2008)
I think this is a RAC competitor, just like Scalable Shared Databases or replicating to slaves.Log shipping existed in many installations before Microsoft built a support solution, this might end up being auto-replication of some sort.
I've heard RAC doesn't sell well, hard to administer, expensive, etc. Perhaps this is way to solve some of those issues as well.
I still keep having visions of some crazy guy who hands me a red pill and shows me that SQL Server was a lie, and that it's actually a simulation running Oracle in the background....
Of course - that WOULD explain the cheap suits and bad sunglasses....:w00t:
----------------------------------------------------------------------------------
Your lack of planning does not constitute an emergency on my part...unless you're my manager...or a director and above...or a really loud-spoken end-user..All right - what was my emergency again?
May 22, 2008 at 10:00 am
In all seriousness though - this just seems to hammer even further on what is usually the weakest link in high volume systems, i.e. the storage subsystem. It seems to me that we're a quantum leap away from a concept like MatrixDB being useful/useable, in that you'd need to radically increase parallelism to your storage (multiple controllers actually able to read from the same areas at the same time, without having to incur the phenomenal overhead it takes to copy every change n times over).
Perhaps I'm a cynic - but the front-end hardware has almost never been the problem when it comes down to performance. The disk sub-system on the other hand has often been involved/part of the problem. And adding more spindles only gets you so far.
I wish them the best - I really hope they find a way to make it all work. I just shudder at the price ticket that would accompany such a system.
----------------------------------------------------------------------------------
Your lack of planning does not constitute an emergency on my part...unless you're my manager...or a director and above...or a really loud-spoken end-user..All right - what was my emergency again?
May 22, 2008 at 12:26 pm
Oracle RAC - How many companies are currently running it? If they are, how many nodes? I am not an Oracle afficiando, but I have heard from some guys who deal with both MS and ORA that RAC is mostly marketing.
Scalable Shared DBs - I have not heard of anyone actually implementing this or even testing it. The big problem is the refresh of the DB that is read-only. I've tried to think of scenarios where this would solve a problem that I have come across or could even imagine, but it is difficult. However, I believe that HP's Polyserve will allow a single instance to write and have other instances attach as read-only.
MatrixDB - Devs will need to where trench coats and shades as well as ride motorcycles to work. Cool name though...
May 22, 2008 at 12:30 pm
Rick Heiges (5/22/2008)
Oracle RAC - How many companies are currently running it? If they are, how many nodes? I am not an Oracle afficiando, but I have heard from some guys who deal with both MS and ORA that RAC is mostly marketing.Scalable Shared DBs - I have not heard of anyone actually implementing this or even testing it. The big problem is the refresh of the DB that is read-only. I've tried to think of scenarios where this would solve a problem that I have come across or could even imagine, but it is difficult. However, I believe that HP's Polyserve will allow a single instance to write and have other instances attach as read-only.
MatrixDB - Devs will need to where trench coats and shades as well as ride motorcycles to work. Cool name though...
Think of it...You could become a SSIS expert in a single 5-second phone call....:cool:
"Whoa....I know OLAP Fu...."
----------------------------------------------------------------------------------
Your lack of planning does not constitute an emergency on my part...unless you're my manager...or a director and above...or a really loud-spoken end-user..All right - what was my emergency again?
May 22, 2008 at 1:18 pm
This just sounds like a natural progression. They already have parallel execution figured out in integration services. Now they just need a good disk partitioning algorithm. Didn't Gamma do round-robin, range, hash, and hybrid-range partitioning? Guess they are all set except for actually writing the code...
Joshua Perry
http://www.greenarrow.net
Viewing 15 posts - 1 through 15 (of 15 total)
You must be logged in to reply to this topic. Login to reply