Performance issues using .NET C++

[TGF]

I am wondering if it is feasible to use .NET for applications that have to
be very fast. We have a few applications that are blazingly fast, written
in Borland C++ using Borland C++ Builder. We would like to port over to
..NET, but some of the test models we have developed are MUCH, MUCH slower
than it's Borland counterpart. Is this due to the .NET Framework and the
JIT? Or is it something we should be looking for in the project settings
and/or compiler settings? The performance decrease is about two-fold.

 

[Guest]

Its probably not the decisive answer you're looking for,
but our guidelines dictate that when performance is
paramount, we go native, typically through ATL7, ATL
Server, OLEDB (IRowsetFastLoad for SQL2K/CBulkRowset and
so on) using custom marshalling where necessary. Where
we need consistency, or where performance or scalability
is less significant we use managed C++. Its probably a
little unfair to expect .Net to compete with native code
for performance, its benefits are most likely in the RAD
arena. This allows us to leverage the skills of our
various in-house people more effectively.
However even C++.Net still offers great benefits over
other managed languages, such as the ability to host
mscorsvr GC, custom-marshalling, etc. These particular
advantages will probably disappear over time, but with
the impending profiling compiler features, provided they
are available to native and managed projects, native
should always beat intermediate in pure performance terms.
I don't wish to provoke jihad (oh, my language is better
than yours because...), but its the same old story as the
arguments between vb/c++, i.e. tools for the job in hand.

HTH,

mk