I have both DBISAM and EDB so i can compare those myself.

My question was kbmMemTable vs EDB local in-memory table speed
difference as per Barry's post?

Raul


On 9/2/2013 12:52 PM, Roy Lambert wrote:
> Raul
> If you can produce a test case I'm happy to convert to ElevateDB and compare. I'll need data and code.

Raul wrote:

On 9/1/2013 2:24 PM, Barry wrote:
>> I could have used a local EDB session and kept the memory local and it would have been faster than a C/S memory table, but it still would not have been as fast as kbmMemTable.<<

>Did you test the speed difference ? I'm curious what makes you think
local EDB in-memory table would not be as fast ?<

I only tested EDB C/S memory table vs kbmMemTable which wasn't a fair comparison because the data was coming from the server using TCP/IP.  (The client was continually waiting for the EDB table data.) At the time I didn't think to create a local EDB memory database.

I have not done a comparison of kbmMemTable vs EDB local memory table, but my money is still on kbmMemTable. I don't have a lot of free time, otherwise I would create a test application and benchmark the two.

Barry

Raul,

<< BTW - Serious question not picking on the post - we use in-mem tables
extensively in our DBISAM app and if we do ever move to EDB we'd want the
EDB in-mem to be as fast as possible (meaning we can ask Tim to investigate
if kbmMemTable is really a lot faster). >>

I don't know about "a lot faster", but it's probably at least a little
faster due to architectural differences.  In-memory tables in EDB are
designed to be shareable between threads like other
tables, so right there you start to get into a whole kettle of fish that
kbmMemTable doesn't deal with.

Tim Young
Elevate Software
www.elevatesoft.com

Tim

>I don't know about "a lot faster", but it's probably at least a little
>faster due to architectural differences. In-memory tables in EDB are
>designed to be shareable between threads like other
>tables, so right there you start to get into a whole kettle of fish that
>kbmMemTable doesn't deal with.

I'd certainly support you on this one, and, having survived the trauma of writing my own custom dataset descendent, I would place a small wager that the GREAT LEVELER is TDataset.

Roy Lambert