Large caches ...

[Hans-Juergen Schoenig]

I know that it is not a database. Basically I fully understand what you 
mean. We have a real database which does the real storage.
The problem is: If we lose a node we cannot reload the entire cache as 
it is too large and the system will be down for quite a while.
 If I know which server got lost - is there a way to repopulate only the 
data which is on this certain server?
This would dramatically reduce the time needed to recover. In the 
meantime the missing data could be provided by the database
(serving 1 out of 10 records from the database should work).

    Best regards,

       Hans


Brad Fitzpatrick wrote:

>You know that memcached isn't a database, right?  It's a CACHE.  If a
>machine dies, you get some cache misses, which you then have to repopulate
>from your REAL DATABASE.
>
>- Brad
>
>
>On Sat, 2 Jul 2005, Hans-Juergen Schoenig wrote:
>
>  
>
>>We are planning to develop an application on top of a very large in
>>memory database. Basically the size of the application will be around
>>100 - 400 gb (in memory). We will need a quite large server farm to
>>achieve that. The problem now is: What happens if a server fucks up?
>>What can we do to make this work 24x7? Rebuilding the cache from scratch
>>is not an option as it is too large :(.
>>
>>How did other folks solve that?
>>
>>    Best regards,
>>
>>       Hans
>>
>>
>>    
>>