Re: Slow initial connection using Simple Channel Access

[Alan Biocca <AKBiocca@LBL.gov>]

Chris Timossi will likely comment further re NT specifics, but just to
give a reference point, these functions make 7,000 successful connections
in something less than 2 seconds from Unix on an old Sparc 2. It is
important to queue all the requests before processing them, do not do
them one at a time. (ie queue everything then one do_get). The
interfaces in the underlying Channel Access are significantly optimized
for batch processing and SCA uses this. In addition, single requests
cause a flurry of packets on the network that can cause tinygrams to
impact network performance for hosts and network neighbors without
getting much work done.

That having been said, 250-300 msec sounds slow even for individual
requests. Possibly a configuration problem.

-- Alan K Biocca

At 09:51 AM 4/9/99 -0500, Dale L. Brewe wrote:
> I've been using the Simple Channel Access
> libraries under NT, and have
> noticed that the time to make an initial connection is quite long.
> For
> example, using the routine cache_getbyname, it takes about 45sec to get
> 140
> values initially (~0.33sec/value). I've also tried the batch
> routines
> que_get and do_get, with a marginal improvement (to ~0.25sec/value),
> which
> is still pretty difficult to live with. Can anyone tell me if they have
> a
> similar experience? Is this an inherent limitation of these
> functions/library, or is it possible to improve the performance? My
> ioc
> uses epics 3.13beta11 and my client (based on LabVIEW) has used
> sca3.dll
> and ca.dll for 3.13.0 and previous versions.
> Thanks
> Dale Brewe
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>  Dale L. Brewe     PNC-CAT/U. of 
> Washington
>  ph: 630 252 0582        
> fax: 630 252 0580
>  Mail:
>  APS/Argonne National Laboratory
>  Bldg 435E sector 20
>  9700 S. Cass Ave.
>  Argonne, IL 60439
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~