[BitTorrent] 0length target file bad performance

dejank dkostic at cs.duke.edu
Sat Aug 28 18:17:27 EDT 2004


I'm running bit torrent in a lab, starting multiple instances on a 
single machine (with virtual IP address handled by the ModelNet 
emulation environment). Everything works well if the target file (--
saves option for the headless script) doesn't exist before starting 
up. If I create the file 0length file, performance suffers 
significantly. I have tracked the different behavior down to code in 
Storage.py that just opens the file with 'rb+' instead of 'wb+' 
(when the file doesn't exist). Barring python bugs, this in itself 
doesn't seem like a smoking gun. I might be missing some detail here.

The ultimate reason I'm doing this is to eventually allow multiple 
BT instances on the same machine to write into the same file. This 
is very important for large-scale evaluations, as I would like to 
look at the performance of the algorithm, and not worry about 
swapping when 20 bit torrent processes are each writing to a 100 MB 
file on a 512MB RAM machine.

Any help would be appreciated, either explaining why downloads don't 
work well if the file exists or hints on getting this multiple 
download a single file experiment going.

Dejan Kostic

------------------------ Yahoo! Groups Sponsor --------------------~--> 
$9.95 domain names from Yahoo!. Register anything.

Yahoo! Groups Links

<*> To visit your group on the web, go to:

<*> To unsubscribe from this group, send an email to:
    BitTorrent-unsubscribe at yahoogroups.com

<*> Your use of Yahoo! Groups is subject to:

More information about the BitTorrent mailing list