1) The product was named Caché because it is very good
at caching, those performance numbers are not from
using the file system cache, and why is using asynchronous
disk I/O an issue for the performance numbers?
The numbers are for sustained activity, not just doing a small burst of inserts and then having the writing occur after the benchmark completes.

2) Triggers are coded in CacheObjectScript, not M[UMPS]
(difference is kind of like C++ compared to C, but even more so)
I believe that triggers can also be written in CacheBasic (like VBScript), or CacheMultiValueBasic (like MultiValiue/Pick Basics)

3) Data in Caché is NOT just stored as text, and does NOT need conversion every time a calculation is done.
They are internally stored as either:
Binary strings,
Unicode strings (but in a much more compact form than either using UTF-8, UCS-2 or UCS-4),
integers,
scaled decimals (64-bit signed #s with a power of 10 exponent from -128 to 127),
or IEEE 64-bit doubles.
Hans may have been confused because to the CacheObjectScript language, those internal types are mostly invisible to the programmer (except for the distinction between normal Caché strings and numbers (i.e. decimal arithmetic) and IEEE floating point values (binary floating point arithmetic)).

You raise excellent points. I hope somebody from the company addresses them.

Thanks,

CAM

That performance might come from disabling transactions and using asynchronous disk writes or the file system cache. Which would put the configuration at exactly how other RDBMS vendors handle BLOBs. Which would explain why anyone would think to compare insert performance of records in one DBMS to that of BLOBs in another.

Having those configuration options is good. But let’s just say that I have been annoyed by the way they phrase various claims in their sales pitch.

They did have a good looking API for doing high speed data inserts via shared memory. But again, something that should be revealed as part of a performance comparison since use of this API involves tradeoffs.

Also, the event system that I saw at that time worked like this:

1) insert event as a record
2) fire the equivalent of a post-insert trigger

The triggers were coded in MUMPS, so it is possible that now they are coded in Java. I sort of liked their trigger mechanism, but it’s not really comparable to what a CEP engine does.

And note that all data in Cache is stored as text. Even numbers, which are automatically converted every time a calculation is done. I’m not judging this, just pointing out an interesting fact.

If you’re into the tech details, it’s fun to learn a about how Cache works. They have very interesting and sometimes unique approaches to handling DBMS functionality.