[Chandler-dev] Re: Test data set for Chandler perf testing

[Philippe Bossut]

Just one idea: Mimi ran an informal survey on the design list some 
months ago asking the number of events, recurring types, tasks, etc... 
each one of us dogfood users have in our various tools. I suppose she 
has a synthetic spreadsheet somewhere with those numbers.

Why don't we use these data to expand the 3000 events test file with 
synthetic data that do reflect a sort of average target user?

Cheers,
- Philippe

Dan Steinicke wrote:
> I looked at the Generated3000.ics file this test uses.  It doesn't 
> look like it is representative of all the event types and has only one 
> type of recurrence (weekly).
>
> Dan
>
> Sheila Mooney wrote:
>
>> Heikki,
>>
>> Sorry, I don't know too much about that 3000 event calendar. How 
>> many  recurring events are there of the various types - weekly, 
>> monthly,  daily, yearly etc. Does it have all-day, @ time, anytime 
>> events? I  think it was put together a long time ago so I would want 
>> to make  sure it's representative of all the stuff we have today.
>>
>> Sheila
>>
>> On Jun 12, 2006, at 1:59 PM, Heikki Toivonen wrote:
>>
>>> Philippe Bossut wrote:
>>>
>>>> May be the simplest would be to take a couple of existing calendars
>>>> (mine, Katie, Sheila, etc...), merge them and use that as a  
>>>> template for
>>>> creating the set.
>>>
>>>
>>> I've been agonizing over this for a while for a few reasons, main one
>>> being that I really wouldn't want to lose the ability to compare  
>>> how we
>>> are doing with the past. If we switch to new data sets that would
>>> happen. And running the performance tests takes a long time, so IMO we
>>> can't really add many more tests either.
>>>
>>> So I am kind of suggesting that we continue using our current 3000  
>>> event
>>> test calendar to run the continuous tests (we identified that we'd  
>>> want
>>> to have 1500 events and 500 tasks which is not terribly far off  
>>> from our
>>> 3k data set). We should still create the 1500+500 data set, and 10000
>>> data set, against which we could do manual runs every now and then.
>>>
>>>> One question I have though: shouldn't the set be recreated before  
>>>> each
>>>> test run with events relative to the current date? Or alternatively,
>>>
>>>
>>> Ideally we would do this. Shouldn't be too difficult, but someone  
>>> would
>>> just have to do it. However, I think this isn't a problem worth fixing
>>> yet, because we are so far away from even our informal targets.  
>>> Once we
>>> get really close to the targets with our current data set would I
>>> recommend working on this.
>>>
>>> -- 
>>>   Heikki Toivonen
>>>
>>>
>>