Re: Wanna try SolidDB for IDS?

On 6 Aug, 13:32, Ian Michael Gumby <im_gu...@hotmail.com> wrote: > On Aug 5, 6:20 pm, "Obnoxio The Clown" <obno...@serendipita.com> > wrote: > > > Neil Truby said: > > > > What particualr circumstances would it be really good for? > > > Because it was designed solely as an in-memory database, the access > > mechanisms and so on are roughly an order of magnitude quicker than IDS. > > So if you want to access rows in microseconds, rather than milliseconds, > > it's your man. > > Actually no. Did you ever look at Kevin Brown's stuff? (Real TIme > Loader)  It loaded the inserts into a shared memory segment that to > the database looked like a table, and could be read as a table until > it was buffered in to the actual database table.  This is how you > could load 25K real time ticks per "node". (Of course its not just the > high speed of loading but that you could actually query said data > too.) This was 1/3 of the Financial datablade bundle that got caught > in the chaos when IBM stepped in. > > > > > > > > The Highlights are described as: > > > > Accelerates Access to Data in DB2 and IDS > > > Achieves Extreme Speed with In-Memory Database Technology > > > Keeps Data Persistent and Recoverable > > > Combines High Availability and Extreme Speed > > > Balances Data Safety, Application Throughput and Recovery Time > > > Lowers costs > > > > But there must be some disadvantages too, as all these technologies > > > involve > > > trade-offs; ie there must be some circumstances where its use would have > > > negligible effect or perhaps even be negative.  Can anyone describe what > > > these are? > > > My best guess would be the price. > > No, go beyond that. Kick the plug out of the wall. Poof! Your database > is gone. > Also if you're bitching about the price, how much does the kit cost > that has enough CPU horsepower, along with buying the memory? > 32GB? 64GB? 128GB of memory? (And then thing about what happens if , > heaven forbid, you have to *gasp* swap? ) > > Of course if you're spending $500,000 (USD) on hardware, 30K for an in > memory DB is cheap.But that's list. > Wait to you see the pricing if you're getting J level discounts to > start with and then negotiate a deal off of that. ;-) > (Ooops! Yes it does happen. ;-) > > > > > > > > > > Basically, it looks quite sexy and I'd like to try it, but at £30,500 per > > > core to licence, I really am going to have a compelling reason even to > > > trial > > > it. > > > Ah! My best guess was a good one! > > > You will also find some other things, such as possibly a slight difference > > in SQL and limits in the functionality it offers. I'm not sure, but I > > think you'd probably lose the option of extensibility, too. So you might > > have to re-code specific applications. > > > I think it's a niche database, a bit like RedBrick was: outstanding at > > what it was designed to do, pretty much useless at being a general purpose > > database. > > > In general, I think if you were out of its target market, using it to > > address performance issues would be much the same as chucking tin at a > > performance problem: it might get you out of it, but it wouldn't address > > the root cause, which would probably just be a missing index somewhere. > > :o) > > No there's more to it than that. > You never defined the "target" market. > I can think of several. Especially when you get in to cloud and grid > computing. > > Maybe that's why you're not seeing an IDS driven TPC-E benchmark. > Maybe they decided to skip IDS and then do one with "Solid". > > Then you'll see Oracle retaliate with their times ten. > > Of course I do wonder since TPC-E is a stock simulation, I wonder why > they don't do the IDS benchmark using the two components of the > financial foundation? > Time Series and Real Time Loader? > > Sorry, there are some applications, where this makes sense. However, > if TimesTen's lessons teach us anything, its not a long term viable > platform on its own.- Hide quoted text - > > - Show quoted text -- Hide quoted text - > > - Show quoted text - "25K real time ticks per "node" is that per second per cpu? How can you see the shared memory segments for the real time loader? Are they included in onstat -g seg output? There is very little information about realtime loader on the IBM Website. http://www-306.ibm.com/software/data/informix/pubs/library/notes/relnotes/datablades.html links to the "TimeSeries Realtime Loader DataBlade Module 1.10.UC2 " release notes at http://publib.boulder.ibm.com/epubs/html/29920160.html So have there been releases since 2005? The latest timeseries as per 306.ibm.com/software/data/informix/pubs/ library/notes/relnotes/datablades.html also appears to be 2005??

On Aug 10, 7:03 pm, "da...@smooth1.co.uk" <da...@smooth1.co.uk> wrote: > On 6 Aug, 13:32, Ian Michael Gumby <im_gu...@hotmail.com> wrote: > > > > > On Aug 5, 6:20 pm, "Obnoxio The Clown" <obno...@serendipita.com> > > wrote: > > > > Neil Truby said: > > > > > What particualr circumstances would it be really good for? > > > > Because it was designed solely as an in-memory database, the access > > > mechanisms and so on are roughly an order of magnitude quicker than IDS. > > > So if you want to access rows in microseconds, rather than milliseconds, > > > it's your man. > > > Actually no. Did you ever look at Kevin Brown's stuff? (Real TIme > > Loader) It loaded the inserts into a shared memory segment that to > > the database looked like a table, and could be read as a table until > > it was buffered in to the actual database table. This is how you > > could load 25K real time ticks per "node". (Of course its not just the > > high speed of loading but that you could actually query said data > > too.) This was 1/3 of the Financial datablade bundle that got caught > > in the chaos when IBM stepped in. > > > > > The Highlights are described as: > > > > > Accelerates Access to Data in DB2 and IDS > > > > Achieves Extreme Speed with In-Memory Database Technology > > > > Keeps Data Persistent and Recoverable > > > > Combines High Availability and Extreme Speed > > > > Balances Data Safety, Application Throughput and Recovery Time > > > > Lowers costs > > > > > But there must be some disadvantages too, as all these technologies > > > > involve > > > > trade-offs; ie there must be some circumstances where its use would have > > > > negligible effect or perhaps even be negative. Can anyone describe what > > > > these are? > > > > My best guess would be the price. > > > No, go beyond that. Kick the plug out of the wall. Poof! Your database > > is gone. > > Also if you're bitching about the price, how much does the kit cost > > that has enough CPU horsepower, along with buying the memory? > > 32GB? 64GB? 128GB of memory? (And then thing about what happens if , > > heaven forbid, you have to *gasp* swap? ) > > > Of course if you're spending $500,000 (USD) on hardware, 30K for an in > > memory DB is cheap.But that's list. > > Wait to you see the pricing if you're getting J level discounts to > > start with and then negotiate a deal off of that. ;-) > > (Ooops! Yes it does happen. ;-) > > > > > Basically, it looks quite sexy and I'd like to try it, but at £30,500 per > > > > core to licence, I really am going to have a compelling reason even to > > > > trial > > > > it. > > > > Ah! My best guess was a good one! > > > > You will also find some other things, such as possibly a slight difference > > > in SQL and limits in the functionality it offers. I'm not sure, but I > > > think you'd probably lose the option of extensibility, too. So you might > > > have to re-code specific applications. > > > > I think it's a niche database, a bit like RedBrick was: outstanding at > > > what it was designed to do, pretty much useless at being a general purpose > > > database. > > > > In general, I think if you were out of its target market, using it to > > > address performance issues would be much the same as chucking tin at a > > > performance problem: it might get you out of it, but it wouldn't address > > > the root cause, which would probably just be a missing index somewhere. > > > :o) > > > No there's more to it than that. > > You never defined the "target" market. > > I can think of several. Especially when you get in to cloud and grid > > computing. > > > Maybe that's why you're not seeing an IDS driven TPC-E benchmark. > > Maybe they decided to skip IDS and then do one with "Solid". > > > Then you'll see Oracle retaliate with their times ten. > > > Of course I do wonder since TPC-E is a stock simulation, I wonder why > > they don't do the IDS benchmark using the two components of the > > financial foundation? > > Time Series and Real Time Loader? > > > Sorry, there are some applications, where this makes sense. However, > > if TimesTen's lessons teach us anything, its not a long term viable > > platform on its own.- Hide quoted text - > > > - Show quoted text -- Hide quoted text - > > > - Show quoted text - > > "25K real time ticks per "node" is that per second per cpu? > > How can you see the shared memory segments for the real time loader? > Are they included in onstat -g seg output? > > There is very little information about realtime loader on the IBM > Website. > > http://www-306.ibm.com/software/data/informix/pubs/library/notes/reln... > links to the > "TimeSeries Realtime Loader DataBlade Module 1.10.UC2 " release notes > athttp://publib.boulder.ibm.com/epubs/html/29920160.html > > So have there been releases since 2005? > > The latest timeseries as per 306.ibm.com/software/data/informix/pubs/ > library/notes/relnotes/datablades.html also appears to be > 2005?? Well thats the thing. Two things killed the Financial Time Series. First the deal with NAG expired, for whatever reason (I've heard both sides of the story from those involved.) Second, IBM at the time, under Janet wanted anything IDS centric gone. In 2005 time frame, there was someone who actually came to a similar conclusion that I tried to explain to both Caldera and Payne about how to make the parsing/subscribing of the incoming ticks easier to maintain. (And ironically it was something I had developed 10 years earlier that was in use at SSB/SMC and then at Software Alliance/ RMT's Radar product. So that would make IBM 14 years behind the curve.) So this is how I knew that there was some sort of "top sekret" project still going on. So in 2005, IBM could sell the RTL, TimeSeries but no NAG. That kind of killed it, but ironically, all you needed was a C/C++ wrapper around a decent financial package. Any package really and you have the Financial Foundation. With Moore's law on your side, you can see smaller hardware giving you the ability to handle all financial transactions occurring in open markets in real time. This opened the door to re-architecting a solution where the database is no longer a sink. However, this market advantage was never really pursued. Oracle became the "logical" choice when replacing Sybase. (But that's another story ...) -G