IMHO to make this practical we need 2 things to happen in the node:

1. A much cheaper (RAM-wise) datastore.

*IF* random replacement is viable, and mrogers' simulations are promising, but 
we need some more simulations with a more realistic model of requestors etc, 
a simple random replacement table would be a good solution.

2. A much cheaper (RAM-wise) client layer.

It should be possible to keep pretty much the entire client queue on disk - 
either in a database, or in flat tables in a reasonably small directory tree 
(we should probably make this pluggable). Then all the node has to keep in 
RAM is high priority requests, and a queue of the next N requests to send for 
each sender, to prevent waiting for disk accesses from causing us to not send 
any requests in a given cycle.

Both of these would be IMHO useful features for 0.7.1. RAM may be cheap, 
however there are a lot of old systems out there, and our RAM usage really 
shouldn't need to be tweaked by users to reflect the size of their queues... 
that's not good for usability. And it'd be great to be able to run on a home 
server type box, even if it's going to be some years before most users are 
able to get a plug-and-play freenet-box. :)

On Tuesday 15 April 2008 19:48, Sven-Ola Tuecke wrote:
> The corresponding AOT binary pre-compiled for mipsel and i486 is ready. Done 
> some fiddeling with -fPIC and friends and now have a working stripped-down 
> libgcj.so.90. Does not need any java runtime. Created some rudimentary use 
> hints in the readme.
> 
> To show off a bit (please apologize) - some characteristics:
> 
> Filesize main prog: 6 Mb (jar is 2,5 Mb)
> Filesize free-ext.so: 4 Mb (jar is 2,5 Mb)
> Filesize libgcj.so: 10Mb (normally 32Mb)
> Startup time: 30sek
>   (300Mhz MIPS, 128Mb RAM, a relative fresh setup)
> Homepage display after: 15sek (complete after 3 minutes)
>   (Freenet startup complete, browser pre-loaded, via ssh tunnel)
> Top says:  virt:33m  res:33m  shr:14m
> Free says: -/+ buffers/cache:      32712      91252
> Utilization: up 2 days,  9:57,  2 users,  load average: 1.73, 0.95, 0.68
> 
> Of course, I expect the memory utilization around 0.3 Gig virtual after a 
> while - java is a memory wastrel as you probably know. And i486 reaction 
> time is ~1/10 or so. Makefile needs rework...
> 
> Download:
> http://download-master.berlin.freifunk.net/sven-ola/freenet/
> 
> // Sven-Ola
> 
> ----- Original Message ----- 
> From: "Matthew Toseland" <toad at amphibian.dyndns.org>
> To: <devl at freenetproject.org>; <support at freenetproject.org>
> Sent: Tuesday, April 15, 2008 6:47 PM
> Subject: [freenet-dev] Freenet 0.7 build 1140
> 
> 
> > _______________________________________________
> > Devl mailing list
> > Devl at freenetproject.org
> > http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl 
> 
> _______________________________________________
> Devl mailing list
> Devl at freenetproject.org
> http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
> 
> 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: 
<https://emu.freenetproject.org/pipermail/devl/attachments/20080416/5e38fc6e/attachment.pgp>

Reply via email to