Simple questions from memcached newbie

Just Marc marc at corky.net
Thu Oct 12 06:23:51 UTC 2006


Hi,

Despite what MySQL would want you to think, cluster is very specialized 
and in its current form will probably only fit very, very specific 
apps.   I don't see it being used as a simple "memcached 
replacement".     I also don't think Cluster's performance bottleneck 
(or rather, bottleneck by design) is at the storage backend, so having 
memcache as a storage backend is not likely to improve performance much.

While I don't agree MySQL Cluster is difficult to configure - you simply 
decide on your setup, create a configuration file and copy it over to 
all your nodes, configuration examples exist around so you don't even 
have to do much.   After reading its docs re configuration it shouldn't 
take you more than a few minutes to do so.

If you really want to use it as a memcached replacement, run it in its 
'diskless' mode so it doesn't write anything to disk on any node, 
everything is then kept in RAM, duplicated, of course.

But from my experience you wouldn't want to go there if you need 
thousands or more queries per second to your cache, if you want to reach 
that kind of performance with cluster you'd have to setup at LEAST a few 
nodes (true, you can setup several nodes per server if they have 
multiple CPUs) but if you need tens of thousands of queries per second 
you'd probably need tens of nodes (or more like 20) and a decent switch 
to push lots of internal traffic between Cluster's nodes (if you have 
lots of writes/updates...).

But as a simple memcache replacement, cluster has significant network 
and other overheads which you may not like, especially in version 5.0 
which is "stable" (but in fact even that version IMHO is not so stable 
nor it is "production grade" -- even if they do run it on many 
production clusters...)
> On Wed, 2006-10-11 at 15:08 -0500, Andy wrote:
>   
>> The difference here is that memcached is at least an order of
>> magnitude
>> easier to setup and maintain than MySQL cluster, and adding an
>> additional "storage backend" on memcached means you need to maintain
>> two
>> sets of software (memcached and mysql cluster) rather than one (for no
>> advantage really, since NDB is a memory based store, just like
>> memcached) resulting in increased complexity.
>>     
>
> Thanks Andy for adding in. I totally missed the maintenance part. Apart
> from performance, memcached is so simple to maintain. Mysql cluster
> requires too much of configuration!
>
>
> Jeetu
>
>
>   



More information about the memcached mailing list