[ghns] ghns status, api, etc.

Frank Karlitschek karlitschek at kde.org
Wed Jun 24 14:50:45 PDT 2009


On 24.06.2009, at 21:02, Josef Spillner wrote:

> Am Mittwoch, 24. Juni 2009 20:44:06 schrieb Jeremy Whiting:
>> I can think of one other way to do this.  The client could ask the  
>> server
>> for a custom feed and give the server all the entry id's it would  
>> like in
>> the feed in the http headers. then get one feed back containing all  
>> the
>> entry information for the requested entries.  That would only  
>> require one
>> http request that way.  What do you think of that?
>
> Sounds good to me and should work up to any reasonable number of  
> installed
> entries. I've thought also about identified (authenticated) users  
> but this
> would trade anonymity for performance, so your suggestion is  
> obviously better,
> even though the system security is slightly worse off when small  
> requests are
> meeting huge replies.
> The world is full of trade-offs, we can't have the ultimate system :)
>

> Josef



Well. This is possible.
The problem with this solution is that we need a specific API call for  
this which means additional complexity which is always evil.
But the bigger problem is that the result of the API call wouldn´t be  
cacheable on the server side because the collection of entries to  
check is quite unique for a specific user. So I have to do a lot of  
expensive calculations on the server.


I would prefere if you simple call the existing content-get method  
which gives you the version of the entry.
This call is very good cachable. I can use an unlimited cache time for  
the entry and simple exire the cache if the developer updates the  
content entry. This would give ma a nearly 100% cache hitrate and zero  
database load.

It´s true that this approach would require more API calls. But the  
calls are much more cheaper than the suggested call.

 From my experience the server load is the limiting scalability  
resource and not the bandwidth on the client. The amount of transfered  
data is not a lot even with my approach.

What do you think?


Cherrs
Frank



--
Frank Karlitschek
karlitschek at kde.org






More information about the ghns mailing list