[RC5] OGR stub sizes
steve at banoffee.demon.co.uk
Sun Feb 20 18:26:29 EST 2000
The original OGR effort allowed you to specify the approximate stub size
you wanted, and would give a start and end stub to check between.
This way you could download a block of any size between one that would
keep a 386 occupied for a day and one that would keep a top of the range
Alpha occupied for a week.
Why doesn't the d.net client/proxy network have any option for this, as
we have had for ages on other contests? I'd always assumed that the
proxies would be given a chunk of stubspace to check and then divide it
up into appropriate portions for the clients, using an appropriate
E.g. (arbitrary unrealistic units)
Proxy tells server "I'm after a block with up to 16 work units in it."
Server tells proxy "Here's stub 1-5-7-4 to 1-5-7-10 with 15.4 work units"
Client tells proxy "I'm after a block with up to 4 work units in it."
[Proxy uses a simple approximation of work in the stub, similar to whatever
was done in OGR-22 and OGR-23 giving something like:
1-5-7-4 : 4.8
1-5-7-8 : 3.9
1-5-7-9 : 3.5
the first stub needs to be split, perhaps up to 60?
Proxy tells client "Here's stub 1-5-7-4 to 1-5-7-4-60, with 4.0 work units"
[1-5-7-4-61 to 1-5-7-10 / 11.4 remains in proxy's buffer]
Since the original OGR effort managed something like this with just a
simple(?) web server script and nothing like d.net's proxy network, I don't
understand why we don't have the facility within d.net's OGR proxy network.
Since many of the comments here about OGR seem to be concerning the vast
variability of work in apparently similar stubs, this type of thing appears
to be needed. While it's true that we can't 'know' in advance how many
nodes there are, a rough calculated estimate would be far far better than
no information at all.
Anyone care to comment?
I hope I haven't sounded too negative in the above.
d.net ID: distributed at banoffee.demon.co.uk
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest
More information about the rc5