workunits/h slow down a lot, after reaching 17900 work unit processed (96%)

What steps will reproduce the problem?

1. I install and configure pyrit for CUDA, benchmark gives 7000 PMK for my GT 
440 plus some 700 PMK for each of the 3 CPU cores.

2. create ESSID and start importing password, total password count 880 Millon 
passwords (I'm using standard filesystem, as I still didn't learn how to use 
MySQL with pyrit (next to do)).

3.Pyrit batch.... the first 48 Hrs computed 17900 Work-units, around 96%, then 
it took another 24 hs to go from 96% to 98% and another day to reach 100%, is 
this normal?, isn't the work-units all of the same size?.

What is the expected output? What do you see instead?

I would expect to have a near-linear speed of x workunits per hour, but after 
reach 96% slowed about ten times or more, but the reported PMKs per seconds 
remains the same (8000-9000 PMKS/s)

What version of the product are you using? On what operating system?

Pyrit 0.4.1 with CUDA compiled and working on Backtrack 5 R2 fully updated, 
Driver is NVIDIA-Linux-x86_64-295.59 installed after disable nouveau.
CUDA toolkit is: cudatoolkit_4.1.28_linux_64_ubuntu11.04

Please provide any additional information below.

Pyrit selftest, get:
All results verified. Your instalation seems OK

also: pyrit check_db: no errors.
and: pyrit verify (which took 6 hours): also no errors
al of this tested when i was at 98%, or 18100 work-units processed from 18295 
total.

This effect also happened exactly the same way in another machine, using ATI 
and CALpp, it was going quite fast until 95% and slowed down, it took 0,5% per 
day! where the first 95% was done in 3 days.

Original issue reported on code.google.com by llapo...@gmail.com on 1 Aug 2012 at 3:55

icelon / pyrit

workunits/h slow down a lot, after reaching 17900 work unit processed (96%) #400