Geometry Pattern Myth

I've heard this one too many times now from too many people. I know
these people aren't stupid, and they're not malicious, they're just
taking for granted information given out by corporate Tech Support and
even in the Help file which is demonstrably incorrect.

The myth which has circulated for many years now is that for large
patterns, Geometry Pattern is faster than non-Geometry Pattern because
using the option disables the parametrics which have to be calculated at
each instance. This has been SW's response when a sheet metal user tries
to perforate a sheet with 1000 hexagonal holes for a ventillation grate
or something like that.

This is documented in the Help file as late as SW2007 sp1. You can find
it in the Help index under Geometry Pattern.

I want to put an end to the myth that the Geometry Pattern option can in
any way be considered a performance enhancing option. It can not. Try it
yourself. I ran tests in 2001 and again in 2007 that confirm this.
Create a 20 x 20 pattern of circular cuts on a flat plate. CtrlQ a
couple times and use Feature Statistics. Depending on your computer, the
rebuild time increases by something in the range of 30%.

The *only* time it speeds up the pattern is when there is an end
condition attached to the patterned feature, such as shown in the help.
In this situation, however, you cannot claim that the option is used for
a performance benefit because it changes the geometry of the result.

In the end, this switch should only be used for the type of feature
shown in the Help, and only to change the geometry, not for performance.
I have ocassionally used the option to make patterns work which would
otherwise not work, but this is not completely predictable (at least I
can't predict with 100% accuracy when it will make a pattern work and
when it won't).

The test was originally conducted to show the difference between sketch
patterns and feature patterns, which is far more dramatic, ranging from
about 300% to 800% increase (feature patterns are far faster than sketch
patterns).

Further, Verification on Rebuild shows approx. a 10x increase in rebuild
time over sketch pattern times, and 0 (that's ZERO) percent increase for
feature patterns. Verification on Rebuild build times seem to increase
exponentially with the number of faces, and depends to some extent on
the part geometry or construction methods. This is why parts with
fillets take a much larger hit than parts with only the big faces
blocked in, and why multibody parts rebuild faster than single body
parts with VOR turned on. (if a single part has 100 faces, VOR checks
10000 (100*100) combinations, but if you have 10 bodies with 10 faces
each, VOR checks 1000 (10*10*10) - its the same number of faces, but
fewer combinations, because faces don't have to be checked between bodies.

Anyway, SW please take this reference out of the help file, Tech Support
please stop spreading this fallacy.

 

Matt,

Good detective work.
I must admit, I never turn on "geometry pattern" unless the feature fails.
Interestingly, you get the impression Solidworks actually know that it is
faster with it switched off, because that is the default setting.....

You repeated what I've heard others say regarding sketch patterns being
slower than feature patterns. Do you know whether this also applies to
"sketch driven patterns"?

Regards,
John Harland

 

I haven't tested that directly, but I would suspect that sketch driven
patterns would behave like feature patterns rather than like sketch
patterns. The important part is the way faces are created. Sketch
pattern creates faces the same way a regular extrude creates them.
Sketch driven pattern creates faces the same way a linear pattern
creates the faces.

 

Thanks for the research, Matt. Now I know I'm not missing much when
"geometry pattern" fails. Might as well stop trying.

 

Geometry patter does, in some instances, have its place as a performance
benefit. I'd not actually tested it since 2006 first came out until you
posted.

Bit of history: When 2006 came out I tested, then after a month or so,
upgraded all our files. Worked fine until I noticed that certain assembly
drawings were taking an inordinate amount of time to rebuild if any change
was made. Tracked it down to pieces of perforated tubing, drawn with actual
perforations ( I now draw plain tubing, annotate, and apply a material to
create the illusion of perforations ). Rebuild times for these relatively
simple parts could take nearly an hour in 2006. I tracked it down to some
sort of funtionality change/bug made to how geometry pattern worked. The
parts had origionally been drawn in 2001-2005 without long rebuilds, so I
had not used geometry pattern. Editing the parts and enabling geometry
pattern in 2006 brought the rebuild times back to what they'd been in 2005.
I submitted it as a bug to my var, who claimed that it was "working as
intended" so would not submit a bug report. Fine, I sent them the remainder
of similar parts that I had and told them to convert them to 2006. Got them
back after a week or so. It was taking 4-5 hours of cpu time to edit each
pattern and enable geometry patterns for a given file.

Looks like 2007 is working similar to what 2005 did. Here is the part
that I just tested ( 2007 sp1 ). Its a 2" diameter tubing with 1330 3/16"
holes in it. It consists of the following features: sheet metal, lofted
bend, cut-extrude1 ( seed hole for linear pattern ), linear pattern1 ( 35
instances along the tubes axis ), cut-extrude2, linear pattern2 ( 36
instances ), and circular pattern1 ( 19 copies of the two linear patterns ).
No holes extend beyond the end of the tubing ( which might cause the
geometry changes that Matt referred to ).

Starting condition is that geometry pattern is checked for all patterns.
VOR is off for all CTL-Q rebuilds. I then unchecked geometry pattern in
each feature, one at a time, and noted rebuild times. I then re-enabled it
for the circular pattern, to double check.

geometry pattern on for all patterns- 49.25, 50.89 seconds
off for linear pattern1- 54.28,53.52
off for linear pattern2- 52.47,51.33
off for circular pattern-62.42,62.06
on for circular pattern-56.52,54.97

Its pretty clear to me that, at least the circular pattern, benefits in
performance from using the geometry pattern option.

I think what is missing from SW is an ACCURATE explanation of what
geometry pattern does, when its applicable, ect. Instead I think they are
giving out information based upon what they would like it to do. It does
not seem to perform the same release to release.

I'd be happy to send Matt the file that I tested, if he wants it. Maybe
he'll see something that I missed. Sorry if this is a bit verbose, after
last years fiasco, patterns became kind of a sore spot for me.

 

Brian,

Yeah, I wouldn't mind taking a look at the parts.

Your post prompted me to take a closer look at end conditions which turn
out to be very important. I used the same part, a rectangular block with
a 20x20 pattern of cut-extruded circle, instances don't intersect with
anything but the flat plate.


no GP GP
offset from surf 2.88 .91
up to surf 2.70 .89
through all .20 .28
blind .08 .27
up to next .08 .28


This says that you pay a big price for using up to and offset from
surface end conditions and that GP shows some performance benefits for
those end conditions, but remember that those are intelligent end
conditions that make each instance of the pattern different, and GP
essentially makes them all the same. (see help for example). So for
these examples, unless the up to and offset from end conditions are
being used when they don't need to be, GP can't be said to produce a
performance benefit, since the finished geometry is different.

In my examples previously I had used the through all end condition,
which suffers about a 30-40% hit by using GP.

What is big news to me is that using blind and up to next offers a 60%
time savings over through all, and take a 300%+ hit from GP. I'm glad I
did this additional test.

So, I believe your results indicate that you may be using one of the
first end conditions. If this is true, you will get a bigger benefit out
of using blind or up to next than you will from Geometry Pattern. An
interesting thing here is that the up to next condition is still an
intelligent end condition, but the speed cost of using it is very low. I
have a new default end condition for patterns, I think.

Can you say what the end conditions are on the patterned features in
your parts?

 

My end conditions are "thru all". I'll send you the part to look
through. But if you really want to see a huge difference, and still have it
loaded, re-create the part in 2006 ( one of the earlier sps ).