On Mon, Sep 08, 2008 at 01:01:43AM +0100, Ed Loach wrote:
> 
> Could perhaps the server stop giving out new requests when active requests + 
> number of requests in update queue > ~95% of maximum upload queue? That way 
> the upload queue is unlikely to fill completely, which seems to be where the 
> problems are at the moment (I've been trying to upload the same result for 
> about 8 hours now, and the intervals between retries are somewhere about 600 
> seconds). 

There's hundreds of active jobs out there are any one time (if the server is 
not broken again :-)). And responses come back sometimes many hours after a 
request has gone out. So stopping the handing out of requests would take a bit 
of time until the system reacts, and I remember horible stuff back from my 
physics classes, with resonance requencies, self reinforcing circuits, leading 
to eventualy breakdowns. I'd rather have a smooth flow than trying to start and 
stop things in this system.

I don't see a full upload queue as a particular problem, it's just another 
bottleneck that needs elimination -- unless of course, it blocks high priority 
requests from being uploaded because the server is busy with empty sea tiles --.

I prepared the server to send out the priorities of a request to the client, it 
just needs a small change and the client will know about it.
Clients can the decide, for example to upload priority 1 requests until the 
upload queue is full. They could hold back priority 4 requests when the upload 
queue is >50% full (or something similar). This way we never block higher 
priority requests

_______________________________________________
Tilesathome mailing list
[email protected]
http://lists.openstreetmap.org/listinfo/tilesathome

Reply via email to