In the mid-90s, there were a lot of predictions that the telco industry was going to take over payments business. The issue was that telco had done a lot of work for high-volume call-record processing ... helping a number of "in-memory" DBMS operations. These defaulted to the data being in memory with periodic checkpoints as opposed data being on disk with in-memory caches (in-memory DBMS claims of ten times performance when compared to traditional RDBMS even when all data was also "cached" in memory).

The prospect was that the looming micropayments volumes could only be addressed by the efficiencies of the telco call-record processing. Then the telcos would leverage that to move up stream and take-over the remaining parts of the payment industry.

Micropayments has been a long time taking off. Also the foreys that some of the telcos had into payments floundered ... frequently because they had a different business model for dealing with fraud.

The intervening years has seen some of those in the payment industry installing "in-memory", ten-times DBMS ... starting to position for much higher payment transaction volumes.

I had worked on original relational/SQL implementation
http://www.garlic.com/~lynn/subtopic.html#systemr

and some of the people show up later at "in-memory", ten-times DBMS startups (although there has been some consolidation with startups being gobbled up by larger/traditional RDBMS vendors).

Hi Ian,

And what do you do if the data does not fit into RAM?

In one application, I have approximatley 200 TB of data. Where do I get that much RAM cheap enough?

Yes, I agree

> And what do you do if the data does not fit into RAM?

then, it doesn't work...

> In one application, I have approximatley 200 TB of data.
> Where do I get that much RAM cheap enough?

That would classify as a silo :)

There's a firm that'll sell you several TB's of RAM in a box with a SCSI interface. You could use that for a swap file !

So what's the difference between a hard disk and a memory module... Is it just
- Write to Hard disk = Slower. Can't lose written data on uncontrolled power loss or process termination. Wide ability to share data with different processes/ servers
- Write to memory = Faster. Can lose written data on uncontrolled power loss or process termination. Limited ability to share data with different processes/ servers

So should the questions be
- How fast do you need your read/ writes to be
- How often do you get uncontrolled power loss
- How often do you get uncontrolled process termination
- How widely do you need to share your data with other processes/ servers