I think what is happening is the way it is streamed into the buffer requires
that the memory it sits in be extended one chunk at a time. I need to figure
out a way to allow the streaming to preallocate all the memory it needs so
that it doesn't have to grow it in the increments.

HTH,
Jason Wharton
CPS - Mesa AZ
http://www.ibobjects.com

----- Original Message -----
From: "Brian Dunstan" <bdunstan@...>
To: <ibobjects@yahoogroups.com>
Sent: Monday, April 02, 2001 10:11 PM
Subject: [IBO] Large Blob Insertion


> Hi,
>
> I am inserting blobs of about 2.5 MB an finding it VERY slow, in the
> vicinity of 3 minutes per blob. The funny thing is it seems to be the
client
> that is slowing the process.
>
> When I post the blob the client cpu utilization goes to about 100% for
about
> 3 minutes while the network remains largely inactive (I have separate
client
> and server computers).
>
> It appears that most of the time is being spent in a procedure called
> IB_ReallocMem in unit IB_Parse. Each time through it is increasing a
memory
> allocation by 4095 bytes all the way up to 11,629,632 bytes which is over
> 2,800 iterations (nearly 4 times the size of the blob). After that is done
> then the network and the server get busy and after a further delay the
post
> is complete.
>
> I have played around with various blob segment sizes and not found any
> dramatic differences. I am currently testing with the maximum of 32,767
> bytes.
>
> Another funny thing is that IB Monitor is reporting a segment size of
65535.
>
> I am using a IBO 3.6Cf and IB 6.01. The components I am using are:
> TIBODatabase, TIBOQuery with RequestLive, TDataSource and a TwwDBRichEdit.
I
> am using the TwwDBRichEdit to insert a OLE BMP object into a RTF document.
>
> My workstation is Win 2000 SP1 with 256 MB and my server is NT 4 SP 6.
>
> Am I doing something obviously wrong here, or is there some parameter I
> should be tweaking to get dramatically better performance?
>
> Are there any native IBO controls that will allow my users to edit RTF and
> insert OLE objects into them with reasonable performance.
>
> Thanks in advance for your thoughts.
>
> Brian