ignite-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From vkulichenko <valentin.kuliche...@gmail.com>
Subject Re: Data grid Load balancing
Date Wed, 10 Feb 2016 19:40:22 GMT
Hi Vinay,

Replicated cache works like a partition cache with backups on all nodes
(plus several specific optimizations). So in replicated cache there is still
primary node for any key and client request will go to that node (i.e.,
affinity behavior is similar to partitioned cache, you're right here).


vinshar wrote
> if above point 2 is correct then, Do you think there will be any gain if
> in case of replicated caches we use round robin or weight based or current
> server load / performance based strategy to decide which server node to
> send a request for a cache operation? or will it be an overkill because in
> case of cache operations (other than get) which changes something in cache
> will anyway will propagate to other server nodes. and it may also be an
> overkill in case if near cache is utilized by clients.

In vast majority of use cases clients request a lot of different keys and
current architecture guarantees that in such cases requests will be spreaded
across different nodes. I believe this is correct default behavior, but you
can always override it using Compute Grid.

Default affinity function is RendezvousAffinityFunction.

-Val



--
View this message in context: http://apache-ignite-users.70518.x6.nabble.com/Data-grid-Load-balancing-tp2819p2938.html
Sent from the Apache Ignite Users mailing list archive at Nabble.com.

Mime
View raw message