FYI: About slow image start-up times

Brian T. Rice water at tunes.org
Sun Aug 15 22:17:50 PDT 2004


This is more of a "helpful info" tip that I realize I hadn't explained 
or documented adequately yet. I'll add it to our documentation in the 
README or elsewhere somehow.

The images that we post on the site for alpha and for release are 
"kernel images": they consist of the absolute minimum framework needed 
to work with the VM, plus one HUGE method that is run on startup and 
consists of /all/ of the library code.

So when you run these images, they "unpack" within themselves to a 
full-sized image, which is why the memory allocator will tell you that 
it's expanding to about 4MB or so (this is additional overhead to allow 
for object movement in memory to free up space when needed). Once you 
save that image, this method no longer exists, and loading that saved 
image will start instantaneously.

Note that this same procedure that we use can also in principle be used 
to create "script" or "executable-oriented" images where the VM just 
runs whatever's in there, which can define as little as it absolutely 
needs. But making this remotely easy to do for arbitrary programs would 
require a lot of additional work and code organization automation that 
we don't have yet. It's something to consider, though.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: water.vcf
Type: text/x-vcard
Size: 208 bytes
Desc: not available
Url : /archives/slate/attachments/20040815/ad3852fd/water.vcf


More information about the Slate mailing list