10-05-04 10:54 PM
How large is the inbound flatfile?
--
Joe Eldridge [MSFT]
This posting is provided "AS IS", with no warranties, and confers no rights.
"Henrik Sandqvist" <HenrikSandqvist@discussions.microsoft.com> wrote in
message news:75F50EBF-8492-44ED-AD25-3777C0DD9284@microsoft.com...
> Hi
> I've run into what looks like a performance/throughput bottleneck in the
> Flat File Serializer component in BTS 2002 when it's generating files with
> large number of records (25000+) in them. My inbound document is an
> XML-file
> with invoicing data (thus the size) and outbound document is a positional
> flat file needed by our legacy system. A small map is also applied in the
> channel.
>
> Example:
> An inbound document with 27.000 records takes 4 minutes to submit into
> BizTalk and the resulting flat file is completed after another 27 minutes
> (!!!) in the work queue. When I remove the envelope spec from the Port
> definition (hence eliminating the Serializer but still using the map and
> creating an XML-file as output) and rerun the XML-file the submission time
> is
> the same but I get my resulting file in less than 2 minutes (15 times
> faster
> than with the serializer).
>
> My only conclusion is that the Flat File Serializer seems to have a
> serious
> performance/throughput problem and it also seems to be linear i.e.
> breaking
> up the files into smaller parts will not speed up the total process or
> resource consumption.
>
> The system used in the above test has a single 2,4GHz P-IV CPU with 512 MB
> of RAM, BTS 2002 SP1. During the work queue processing BizTalk server
> consumes 100% of the CPU but there is plenty of RAM available. No other
> messages were processed by the server during testing.
>
> Has anybody experienced something like this? Is there an explanation
> (maybe
> from Microsoft)?
> We're targeting 200-400.000 records/day and letting the files sit in
> BizTalk
> spending CPU-hours (!) is not an attractive solution. Writing a custom AIC
> just to produce a flat file is not a very good option either, in my
> opinion
> this should work in the shipped standard component with acceptable
> performance, I don't demand a lightning fast solution but definitely one
> better than this.
>
> Regards
> /Henrik Sandqvist
> Resco AB
[ Post a follow-up to this message ]
|