Spectre does not use new netlist

A while back, there was a post about spectre (via ADE in interactive
mode) not using the new netlist when changing component values on the
schematic. The workaround was to use spectre in batch mode,
resulting in longer simulation load times when only changing variables
in ADE.

The fix:

Check to see that there is only one entry per license file in the
environment variables LM_LICENSE_FILE or CDS_LIC_FILE. ex:
CDS_LIC_FILE={path}/cadence_license
LM_LICENSE_FILE={path}/cadence_license:{other license files}
For speediest performance, unset LM_LICENSE_FILE. Even though the
docs claim that CDS_LIC_FILE has precedence over LM_LICENSE_FILE, we
found that the latter is still parsed when icfb loads.

The length of CDS_LIC_FILE/LM_LICENSE_FILE also affects how long it
takes icfb to load. In my case, there was an order of magnitude
improvement for the time it takes the CIW window to come up. Another
positive side effect is that ADE->Simulation->Netlist and Run (with
spectre) is be MUCH quicker.

The why:

The cause of netlist problem has to do with license checking causing
interprocess communication timing problems between spectre and ADE.
Depending on how/where one's environment variables are set,
LM_LICENSE_FILE or CDS_LIC_FILE may have duplicate entries for license
files. ex:

LM_LICENSE_FILE={path}/cadence_license:{path}/synopsys_license:{path}/cadence_license:{path}/synopsys_license:{path}/cadence_license:{path}/synopsys_license

After much debugging, we discovered that Cadence spends time checking
through EVERY license file that is listed in the environment variables
even though the Cadence license file is listed first. This happens
for IC446, IC5.0.33 on both solaris and linux platforms.
The time taken to parse every listed (duplicated!) license file makes
many parts of cadence slow to load (spectre, icfb, ocean scripts)

This slowness is exacerbated if both CDS_LIC_FILE and LM_LICENSE_FILE
are set and a non-cadence license file which has many features is
listed multiple times (in my case, another tool with over 400
entries). I was waiting upwards of 20 seconds before spectre would
even *start running* after 'Netlist & Run' on a ~20 transistor
schematic (as a single user on a linux dual 3GHz cpu, 2Gig RAM
machine!). Now, the same simulation FINISHES in under 10 seconds.

I'm not sure why Cadence checks the other license files once it finds
its own. (or even whether this is a feature or a bug!)

Dee.

 

Dee:
We have 4 license paths in $CDS_LIC_FILE. I have seen the case where
one license server goes down and it takes 5 minutes for icfb to come
up. Have you opened a SR with Cadence to improve the license file
parsing?

 

Hi guys, thanks for all the suggestions! I've talked to a bunch of other
grad students at other schools and this is a real problem with Cadence and
the Analog Environment. The license files were suspects. For now I'm going
to try out all the suggestions and see what happens.

Marc

 

Erik,

I submitted two service requests related to this (SR 32893732: Spectre
takes a long time to load) and (SR 32893816: ADE not simulating latest netlist)
They were quick to respond, unfortunately, their suggestions were not too
helpful.

After we discovered that it was licensing that was causing the problems
I let them know that the two requests were related.

Dee.