[RC5] RC5-72 buffer files and old assigned blocks

Elektron elektron_rc5 at yahoo.ca
Tue Jul 13 10:50:08 EDT 2004


On 13 Jul, 2004, at 03:38, Andrew Manore wrote:

> I found buffer files for RC5-72 and OGR on another computer a few 
> months
> ago and decided to upload the finished units from the buff-out files. 
> If
> I remember correctly, these files were dated from around January of 
> 2003.
>
> I've processed a good portion of the 330 or so work units that were in
> the old buff-in file, and have 212 work units remaining - 1.20:41:48.00
> worth of work on my system (P4 Prescott @ 2.8ghz, 800mhz FSB).
>
> I'm wondering if I should go ahead and finish the rest of the work 
> units.
> Does the RC5 projects proceed in a linear fashion, or are units taken
> from random parts of the key space and distributed? Basically, I'm
> wondering if I'll be doing work that's already been completed. 
> According
> to the most recent statistics, I'm apparently receiving credit for 
> these
> blocks.

You receive credit for work that's already completed. Except, as far as 
I know, DNet goes through all the blocks before it starts reissuing 
blocks, so already completed work should be no issue (unless it runs 
out of blocks and thus picks a random block).

You could also use dnetc -import, and then just leave it:
-import <filename>   import packets from file <filename> into client 
buffers

> does anybody know EXACT algorithm by which is Dnet server working with
> blocks? I mean, in theory, server must give to us blocks by linear
> algorithm and only in the END of all work server should send us blocks
> which were sent in the beginning and were not completed till those
> day..

Having a purely linear algorithm would defeat the purpose of having 
proxies, since the proxies would need to know which blocks have been 
distributed by other proxies. As far as I know (I'm just guessing), the 
master distributes "superblocks", and the proxy servers distribute 
smaller blocks. From experience with RC5-64, if you want larger blocks 
(for larger effective buffers), launching a few dnetc processes works 
well, since it downloads blocks faster, preventing other clients from 
disrupting your big blocks with smaller blocks.

> But in practice I had some troubles when my completed blocks weren't
> send to server by several month.

I think this has happened in the past due to proxies going down, or so. 
Small flushes may be better, since if work is lost, not much is. I 
flush daily (or so), so I wouldn't know if completed blocks were lost.

123



More information about the rc5 mailing list