64-bit Allegro/Windows vs 64-bit LW/Mac

[Dave Cooper]

Machines under my control recently submitted two test results from 64-bit implementations: 64-bit LispWorks/MacOS, which finished in 4504 seconds, and 64-bit Allegro/Windows, which finished in 44787(!) seconds, almost a full order of magnitude longer. 

Clearly there is something pathological going on with the 64-bit Allegro/Windows. I'm a bit strapped for time at the moment to dive into these results, but maybe someone else has the time and interest to do so? 

My Windows configuration is Windows VirtualBox guest on a mid 2011 iMac 3.4GHz Intel Core i7 (Mavericks) host. This VM is configured with 4 processors and 5GB RAM. The Mac configuration is native MacOS Mavericks on the mid-2011 iMac 3.4GHz Intel Core i7. 

The difference between native and virtualized machine should be nowhere near an order of magnitude. Virtualbox claims "near-native" performance when running the guest in the same architecture as host which I believe I'm doing in this case (X86_64). 

The main thing I noticed while the tests were running for 64-bit Allegro/Windows was that many, many of them (e.g. lispbuilder, opengl stuff, and many more), would simply time out using up the maximal 15 minute period. I'm pretty certain this is what gobbled up the lion's share of that 44787 seconds. But at this point I have no clue what was causing all those timeouts. 

[Anton Vodonosov]

(float (/ 44787 (* 15 60))) => 49.763332

where 44787 is the number of seconds taken by your ACL 64-bit test run,
and 15 minutes is the timeout.

At most 50 timeouts happened, out of almost 1000 libraries,
having around 1500 ASFD systems in total.

It's not enough to make conclusions of course, further investigation is needed.