From: Anne & Lynn Wheeler on

"Steven G. Johnson" <stevenj(a)> writes:
> What you're referring to is that at one point we found performance
> improvements (I believe on an IBM RS/6000, if I remember correctly),
> by inserting padding into the middle of a *one-dimensional* array
> (again to avoid cache conflicts), but it didn't seem like there was a
> sane interface for specifying a 1d array with padding in the middle.

we were brought in to look at issues with one of the major airline
res. systems .... first looking at "routes" (basically finding flt(s)
from origin to destination) ... that accounted for 25% of workload on
large collection of mainframes.

i redid the paradigm and implementation on rs/6000 ... as part of trying
to get at least ten times performance improvement. Initial pass got
twenty times performance improvement ... then some careful tuning of
cache line considerations got another factor of five times improvement
(100 times improvement overall). I added in bunch of new features
(including collapsing several human interactions into single operation)
.... which then brought overall performance of that single interaction
back down to about ten times (but it was eliminating several additional

40+yrs virtualization experience (since Jan68), online at home since Mar1970
First  |  Prev  | 
Pages: 55 56 57 58 59 60 61 62 63 64 65
Next: Texture units as a general function