On May 11, 2009, at 3:23 PM, Matthew Toseland wrote:

> On Monday 11 May 2009 18:15:52 Robert Hailey wrote:
>>
>> Perhaps I am going crazy, but I seem to recall a statistic with
>> regards to the end-result of requests (when Matthew had me tracking
>> down the FATAL_TIMEOUT) errors.
>>
>> Something like this:
>> +-----------------------+
>> |   SSK final states    |
>> +-----------------+-----+
>> | FATAL_TIMEOUT   | 12% |
>> | REQUEST_TIMEOUT |  6% |
>> | SUCCESS         |  2% |
>> +-----------------+-----+
>>
>> Where is it? has it ever existed on the stats page?
>>
>> Surely if 80-95% of requests are failing, we should know why they are
>> failing... right?
>
> They are mostly failing because of DNF because SSK requests are all  
> polling
> nonexistent keys, surely?

That makes sense, but is it also DNF for CHK requests?

Surely the vast majority of CHK requests correspond to data which has  
been inserted, and such a table might show the effectiveness of bloom  
filter sharing with a new final state "BLOOM_FOUND" (or the like).

> We had such a table for swaps, and we have something similar for  
> backoff
> reasons, but I don't think we have anything like that for requests.

I presume what I was looking at much earlier was on the peer list for  
backoff, thanks.

--
Robert Hailey


Reply via email to