Do XPS MS lead to fatal simulation errors?

I ever used IC6.1.6 version, and spectre or APS is so slow if the simulation scale is large. Now I switched version to IC6.1.8 and I found XPS MS. After a few trails I am sure this simulator is mucher faster.

However, when I used XPS MS in a large scale, mix-signal simulation, it always abort. The CW said: 

ERROR (ADEXL-5011): While simulating run Interactive.31, point 1, test xxx:1, received error:
Simulation Error:
------------------------------
Simulator failed to complete the simulation.
FATAL (SPECTRE-18): Segmentation fault.

and simulation log said it stores saome data after the corruption.

And before I tried to relaunch simulation even if I changed nothing, an error will arise:

ERROR (ADEXL-1708): Simulation run failed because the following test-associated cellviews have been modified since their last extraction:
xxx:1}

And I have to check hierarchical though I changed nothing.

My setting is: aps++ on, liberal interval, auto multi-threading, digital speed 2 with digital modules correctly assgined, DC OP solver is MS, and the 100us simulation would always abort at about 60~70us.

And I found that if I used too many hierarchies in simulation schematic, the signals that lied within lower hierarchies might be simulated uncorrectly or totally wrong.

So I don't know which lead to this, is this caused by aps++ mode, digital module assignment, software version, or just problems with the linux itself?