[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: What price security?
>Don't know about your tests, but when I looked at this, almost all
>crypto timing tests were influenced heavily by cache effects, meaning
>that actual mileage is usually much lower than the tests indicate,
Well....yes. But I'd expect the effect on cache hit ratios from task
switching to be fairly minimal except under very unusual
circumstances. The inner core of a cipher is usually pretty small,
and it probably doesn't take very long after a task switch (or after
execution begins, for that matter) for all the important bits to get
loaded into the cache. So unless the task switch rate is
extraordinarily high, I wouldn't expect the effect to be very
significant. IDEA is probably better than DES in this regard, as it
lacks DES's lookup tables.
Furthermore, level 2 caches are getting pretty big these days, to
where I'd expect much of the cipher code and data to still be there
after a brief digression to another process. (If the cipher task is
suspended more than briefly, then it doesn't matter as much
Also, my benchmarks were taken on a system that, while otherwise idle,
still had a clock ticking away at a fairly rapid clip.
Phil
References: