Dave


I'm posting here cos I forgot to flag the other post.

My results are below. Interestingly my slower machine performed better and I'm guessing its down to its being newer so faster hard drive. The amount of memory MAY make some difference as may the processor but I'm betting the disk is the main thing. Also Tim works through OS calls and I have no idea what the other engines are doing but I bet Paradox is doing entirely its own thing.


DBISAM 4.25 Build 5
ElevateDB 1.05 Build 2


HP Pavilion PIV 3.2Ghz 60Gb disk 1GB RAM
WinXP SP2 Home
Partition run on 10Gb of which 5.7Gb free

DBISAM Elapsed: 1,846.5 sec, #Rcds:1,000,000 542 Rcds/Sec Trans Flush: No Trans Size=1500 Build Index Before
ElevateDb Elapsed: 980.5 sec, #Rcds:1,000,000 1,020 Rcds/Sec Trans Size=100 Build Index Before
Paradox Elapsed: 17.3 sec, #Rcds:1,000,000 57,710 Rcds/Sec No Trans Build Index Before

DBISAM Elapsed: 504.7 sec, #Rcds:1,000,000 1,982 Rcds/Sec Trans Flush: No Trans Size=1500 Build Index After
ElevateDb Elapsed: 307.5 sec, #Rcds:1,000,000 3,252 Rcds/Sec Trans Size=100 Build Index After
Paradox Elapsed: 34.2 sec, #Rcds:1,000,000 29,211 Rcds/Sec No Trans Build Index After

DBISAM    Elapsed: 57.3 sec, #Rcds:1,000,000  17,453 Rcds/Sec  Trans Flush: No Trans Size=1500 No Index Built
ElevateDb Elapsed: 77.1 sec, #Rcds:1,000,000  12,963 Rcds/Sec  Trans Size=100 No Index Built
Paradox   Elapsed: 17.0 sec, #Rcds:1,000,000  58,665 Rcds/Sec  No Trans No Index Built

Gateway
Celeron 520 1.6 Ghz
80Gb Hard drive 53 Gb free
Vista Home Basic (aero mainly switched off)

DBISAM Elapsed: 188.6 sec, #Rcds:1,000,000 5,303 Rcds/Sec Trans Flush: No Trans Size=1500 Build Index Before
ElevateDb Elapsed: 284.3 sec, #Rcds:1,000,000 3,517 Rcds/Sec Trans Size=100 Build Index Before
Paradox Elapsed: 14.1 sec, #Rcds:1,000,000 70,987 Rcds/Sec No Trans Build Index Before

DBISAM Elapsed: 217.8 sec, #Rcds:1,000,000 4,591 Rcds/Sec Trans Flush: No Trans Size=1500 Build Index After
ElevateDb Elapsed: 243.5 sec, #Rcds:1,000,000 4,106 Rcds/Sec Trans Size=100 Build Index After
Paradox Elapsed: 34.0 sec, #Rcds:1,000,000 29,432 Rcds/Sec No Trans Build Index After

DBISAM    Elapsed: 60.1 sec, #Rcds:1,000,000  16,641 Rcds/Sec  Trans Flush: No Trans Size=1500 No Index Built
ElevateDb Elapsed: 78.9 sec, #Rcds:1,000,000  12,671 Rcds/Sec  Trans Size=100 No Index Built
Paradox   Elapsed: 17.6 sec, #Rcds:1,000,000  56,676 Rcds/Sec  No Trans No Index Built


Roy Lambert

Roy,
  The interesting thing about your experiment is DBISAM is 10x faster
on the slower Vista box (build index before) than your XP box, and EDB
is 3x faster. Does that mean Vista has better disk caching? I think the
problem is your HP box is really slow, much slower than my benchmarks on
a 4400x2. Maybe your XP box needs defragging? If you have the time,
delete the data files on XP, defrag, and benchmark just the times below.
Your XP box shouldn't be that slow.

Dave


> HP Pavilion PIV 3.2Ghz 60Gb disk 1GB RAM
> WinXP SP2 Home
> Partition run on 10Gb of which 5.7Gb free
>
> DBISAM Elapsed: 1,846.5 sec, #Rcds:1,000,000 542 Rcds/Sec Trans
Flush: No Trans Size=1500 Build Index Before
> ElevateDb Elapsed: 980.5 sec, #Rcds:1,000,000 1,020 Rcds/Sec Trans
Size=100 Build Index Before

> Gateway
> Celeron 520 1.6 Ghz
> 80Gb Hard drive 53 Gb free
> Vista Home Basic (aero mainly switched off)
>
> DBISAM Elapsed: 188.6 sec, #Rcds:1,000,000 5,303 Rcds/Sec Trans
Flush:    > No Trans Size=1500 Build Index Before
> ElevateDb Elapsed: 284.3 sec, #Rcds:1,000,000 3,517 Rcds/Sec Trans
> Size=100 Build Index Before

Dave


To me the interesting point is that it indicates its down to disk speed almost entirely. If you look at your earlier post you had ElevateDB taking c58secs (you'd missed out the index build) mine comes out c80secs. So I presume processor speed and RAM don't have much of an impact.

Just had a look in Norton Speedup and the E: drive I used isn't badly fragmented, but it is a partition on a 60Gb drive with only about half its space left blank. The PC is 3 years old. The CPU (twin core 3.2GHz PIV) still has a lot of punch but obviously the drive is a bit slower than a modern machine.


Roy Lambert

Dave


About as I expected

ElevateDb Elapsed: 884.2 sec, #Rcds:1,000,000   1,131 Rcds/Sec  Trans Size=100 Build Index Before

I think tomorrow I'll have a look at the disk to see what speed etc. - the pc's been away for repair. I pulled the disk cos I didn't want my accounts etc subject to their scrutiny so they can't have damaged the disk and memory tells me it was faster then the Vista pc before.

Roy Lambert

Dave,

<< The interesting thing about your experiment is DBISAM is 10x faster on
the slower Vista box (build index before) than your XP box, and EDB is 3x
faster. Does that mean Vista has better disk caching? I think the problem is
your HP box is really slow, much slower than my benchmarks on a 4400x2.
Maybe your XP box needs defragging? If you have the time, delete the data
files on XP, defrag, and benchmark just the times below. Your XP box
shouldn't be that slow. >>

I think the whacky times are all down to issues with the benchmark becoming
disk-bound.  You could probably cause EDB to be disk-bound if you moved the
transaction size to a very small size, but DBISAM tends to become disk-bound
a lot easier due to the larger amount of index pages being written with
every commit.

--
Tim Young
Elevate Software
www.elevatesoft.com

Dave


I've had a look at the drive - Hitachi 4200rpm - doesn't mention cache. I'll have to find a bigger faster one

Roy Lambert

Roy Lambert wrote:
> Dave
>
>
> I've had a look at the drive - Hitachi 4200rpm - doesn't mention cache. I'll have to find a bigger faster one
>
> Roy Lambert
>

D'uh??! 4200 RPM? Did it come with its own squirrels, or did you have to
catch your own?

Dave