Dustin: Is there not a step like, ``if it's not cached, we need to
look it up?'' If not, then it seems like you want something more
along the lines of dynamo/chubby/zookeeper for these items.
Dustin In this particular case to do that would completely negate the
benefit of the cache because better than 99% would fail in the cache
lookup AND the database lookup. In effect we would be using the cache
as a virtual table and only refer to the database on startup. There is
method in our "madness" uhahaha. The number of "rogue" actions
requested that would be blocked may only represent 10% of our total
action requests, but would reduce server load significantly.
I take the point of some folk who have suggested running a 2nd
instance of memcached and I am thinking about that. The single
downside is calculation of the memory requirement and getting it right
so that we don't lose entries and don't waste memory.
I've not heard of dynamo etc and will look into them.
regards
Grant
On 12/06/2008, at 2:37 PM, Dustin Sallings wrote:
On Jun 11, 2008, at 18:02, Grant Maxwell wrote:
If the class is cached and the record exists then we allow the
action to proceed.
If the class is not cached then we just allow the action to proceed
if the class is cached but the record does not exist then the
action is denied.
Is there not a step like, ``if it's not cached, we need to look it
up?'' If not, then it seems like you want something more along the
lines of dynamo/chubby/zookeeper for these items.
The proposals that make sense to me around LRU policies are more
for QoS kinds of things. That is, you can rate something between
``this is a little cheaper than recomputing it, but whatever'' to
``for the love of [deity] do not uncache this before the expiration
date.'' Even in the latter case, you must accept that the value
*may* not be where you think it is and will need to be recomputed.
The only time I really see making a business decision based on what
value is in the cache is if you're under a heavy load scenario and
are really trying to walk on eggshells around your centralized
resource. e.g. you might give an HTTP 503 kind of response, but
definitely nothing in the 400s.
--
Dustin Sallings
