[GNUnet-developers] Re: [Help-gnunet] Major problems on RH7.2

[Christian Grothoff]

On Sunday 05 May 2002 12:02 pm, you wrote:
> If you agree to accept the gdbm patch that someone sent you,
> it should fix that issue half-way, if you decide to always
> ask the key from the database instead of caching them in
> the memory. Its certainly much faster than asking for
> the file from the fs. Anyway, there'll be concurrency control
> issues probably, as gdbm accepts only one writer at a time.
> The patch supplier might have already taken that into account,
> so I'm just speculating.

The patch does not seem to really go deep into concurrency --- yet :)

> All in all, gdbm seems a much better solution (well, *anything*
> seems) than storing 1kb pieces as separate files. I agree
> to those claiming that not many are willing to use obscure
> filesystems just for gnunet.

Well, no. The gdbm patch will not solve the issue of GNUnet not seeing new 
content instantly because it arises in the existing 
triple-hash-to-double-hash database (yes, another one). This database is 
opened r/w by gnunetd. If gnunet-insert does a concurrent write, gnunetd does 
always get to see the changes... 

> ps. as the gnunet's traffic shaping is what it is,
> I attempted to do an UDP upstream limiting script with tc
> and qdisc. I'm not entirely sure yet if it works 100%,
> but atleast it can, in theory, be done that way, and
> then set gnunet.conf to full blast. I'll send you the
> patch to be put to contrib/ or whatever, if you like.
> I must first make sure though that it approximately
> does what its supposed to do.

... and please document how it works... :-)

Christian