Jim Gray - Part II of talking about Database Design

Awesome stuff here.  His humility (focusing on helping others to succeed rather than on accomplishments linked totally to himself) and love for the subject is inspiring.

Regarding the 2 vs. 3 tiered model, I seriously think the security will not be worked out and thus 2-tiered will, as he said, be a non-starter.  Why should an outsider be forced to use wssecurity?  Furthermore, eliminating the layer between the world wild web and the DB could be a huge problem.  Imagine bringing all the exploits of IIS (and Apache) right to the DB with sensitive data...

His thoughts about asynch vs. synch programming are very useful and timely (I'm doing some concurrency stuff right now for work while I'm still in college, and it's not easy).  Interestingly, I'm working with C# and am noticing the idiosynchrasies of GUI/asynch communications and dealing with passing of messages.

Regarding the Google Labs paper he mentioned, it is here...

What would be cool is if you could somehow get Jim Gray and Jean Paoli in the same room to discuss XML in next-generation OSes.  And maybe thrown in Don Box as a bonus.

In Terra Server, when one of these 'bricks' fails, how long does it take to replicate the working half of the mirror to the hot spare server? I mean, if they've got 3Tb on each machine and they're using probably gigabit ethernet to connect them, that's quite a big window where you only have one copy of the data active. OK, so within each brick the data is mirrored using raid, so even if both servers containing a given data set went down it's likely the data would survive, it would result in a loss of service until the machines could be restored.

Google has a lot of information on their site about how they cope with the same sorts of issues Jim describes. And it is a whole layer above the tranditional OS, so they are not only using commodity hardware, but also a commodity OS. Their GFS filesystem is called a filesystem, but the OS knows nothing about it. It's really a set of libraries and daemons that run entirely in userspace and use plain old sockets for communication. It seems like OS developers have really missed something here as while GFS gets the job done, it's only available to programs written to the interfaces in those libraries.

Regarding the 2 vs. 3 tiered model, if it was an enterprise solution where everything is behind the firewall would the security issues be lessened and therefore the 2 tier might work.

Thanks much.  I would really to hear or see more on his async/sync ideas and patterns.  Any papers/links from Jim on this?  TIA

--William Stacey[MVP]

Although Don Box is my personable dude and savior, I'm going to have to with Jim Gray on this whole preference for sequential programming thing.

Man!! twice i've attempted to download the last part but each time i lost power in the entire house!

going to try again tonight!!

from the looks of the feedback it's going to be great!

Alex Kazovic wrote:

Regarding the 2 vs. 3 tiered model, if it was an enterprise solution where everything is behind the firewall would the security issues be lessened and therefore the 2 tier might work.

Just because its exclusively an internal app, doesn't really mean you have less security threats. Remember, every computer on your internal network is a potential threat. Especially when users have access to the internet and may get infected with a virus.

If you enable Windows Authentication with either the 2 or the 3 tier models, then you'll have the best protection for restricting access to the different tiers.

I don't see much difference with the 2 tier model when stored procs are being exposed as web services directly from the database, compared to having a seperate web service that connects to the database. The only thing different is that you have the web service tightly integrated with the database illiminating the overhead of having the web service connect to the database.

Very cool feature of SQL Server!

Although Don Box is my personable dude and savior, I'm going to have to with Jim Gray on this whole preference for sequential programming thing.

I would also agree.  A lot of current examples and work goes on with async, but anyone that has done an async server beyond something really simple knows the pain involved with async/states and getting it *right.  I would like to hear/read more on how he virtualizes async in a sync wrapper or what ever his pattern(s) are.  I also agree on his xml/rpc notion.  However, the one way that using xml documents instead of RPC/interface contracts is better is that the document can change or be added to.  And if your consumer can understand the new fields in the document it can use them or just use the ones it understands.  So more flex there.  With RPC contracts and no xml, you would have to do a new interface(s).  I love his thoughts on tiers.  I have thought this for some time and was thinking I was crazy for not being a big three tier fan.  Cool to see someone like Jim has been thinking this also.  That said, one could argue that it is still 3-tier.  It just that the middle tier is now in the SQL server web service instead of on another box.  But this is great to have it much easier to reason about and work on.  SQL 2005 should be off-the-hook for developers.  Get ready to hear the wines from DBAs for some time however   I posted a sample sync server in the Sandbox after reasoning about some of Jims thoughs.  Maybe Jim and/or some of his folks could post some written works on these thoughts before they get lost with time and people.  Cheers.

   

I just picked up a great article from /.:

A Call to Arms - Long anticipated, the arrival of radically restructured database architectures is now finally at hand.