On Thursday 03 July 2008 08:21, Michael Rogers wrote:
> Matthew Toseland wrote:
> > - Memory usage decoupled from queue size: The db4o branch "pages in"
requests
> > on demand, so in theory can run with a relatively low memory limit, even
with
> > a very large queue. In practice this would involve a lot of disk I/O, so
it
> > may not work so well in practice, but at least it will degrade gradually,
> > rather than getting OutOfMemoryException's!
>
> With three large downloads running, the db4o branch consistently uses
> about 50-70% CPU (I guess this is the paging you mentioned). The main
> branch uses 90-100% during startup but only 5% after that.
:(
Optimisation is in progress. A lot of work has been done already (especially
yesterday), the next step is to separate the onGotKey() API from the main
SendableRequest (queued requests) interface, so that we almost never need to
touch pendingKeys; at the moment, access to pendingKeys is a significant
proportion of total time spent on the database thread. On my test node, with
a single big download running, the node's normal priority queue grows
indefinitely ("StatisticsToadlet.databaseJobsByPriority", priority 5).
Granted that's with heavy logging...
>
> Cheers,
> Michael
-------------- 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/20080704/c26edafd/attachment.pgp>
