Is replication asynchronous ?

Question

I have a client with a Near Cache, and with the configuration specified as "Client Configuration" below, and two servers with the configuration specified as "Server Configuration" below (note: added the backup-count node). When an object is put into the cache on the client side, it is reasonable to expect the object to be synchronously sent to one of the servers.
The question is, does the replication of that object to the second server get done synchronously or asynchronously? Is there a way to explicitly specify that?
Thanks
Adsen
Client Configuration:
<?xml version="1.0"?>
<!DOCTYPE cache-config SYSTEM "cache-config.dtd">
<cache-config>
<caching-scheme-mapping>
<cache-mapping>
<cache-name>*</cache-name>
<scheme-name>near-scheme</scheme-name>
</cache-mapping>
</caching-scheme-mapping>
<caching-schemes>
<near-scheme>
<scheme-name>near-scheme</scheme-name>
<front-scheme>
<local-scheme>
<scheme-ref>local-cache</scheme-ref>
</local-scheme>
</front-scheme>
<back-scheme>
<distributed-scheme>
<scheme-ref>distributed-cache</scheme-ref>
</distributed-scheme>
</back-scheme>
</near-scheme>
<!--
Default Distributed caching scheme.
-->
<local-scheme>
<scheme-name>local-cache</scheme-name>
<service-name>LocalCache</service-name>
<eviction-policy>LRU</eviction-policy>
<high-units>0</high-units>
<low-units>0</low-units>
<unit-calculator>FIXED</unit-calculator>
<expiry-delay>0</expiry-delay>
<flush-delay>0</flush-delay>
<pre-load>false</pre-load>
</local-scheme>
<distributed-scheme>
<scheme-name>distributed-cache</scheme-name>
<service-name>DistributedCache</service-name>
<backing-map-scheme>
          <local-scheme>
          <scheme-ref>unlimited-backing-map</scheme-ref>
          </local-scheme>
</backing-map-scheme>
<autostart>true</autostart>
</distributed-scheme>
<local-scheme>
     <scheme-name>unlimited-backing-map</scheme-name>
</local-scheme>
</caching-schemes>
</cache-config>
Server Configuration:
<?xml version="1.0"?>
<!DOCTYPE cache-config SYSTEM "cache-config.dtd">
<cache-config>
<caching-scheme-mapping>
<cache-mapping>
<cache-name>*</cache-name>
<scheme-name>distributed-cache</scheme-name>
</cache-mapping>
</caching-scheme-mapping>
<caching-schemes>
<!--
Distributed caching scheme.
-->
<distributed-scheme>
<scheme-name>distributed-cache</scheme-name>
<service-name>DistributedCache</service-name>
<!-- To use POF serialization for this partitioned service,
uncomment the following section -->
<!--
<serializer>
<class-name>com.tangosol.io.pof.ConfigurablePofContext</class-name>
</serializer>
-->
<backing-map-scheme>
<local-scheme>
<scheme-ref>unlimited-backing-map</scheme-ref>
</local-scheme>
</backing-map-scheme>
<autostart>true</autostart>
<backup-count>1</backup-count>
</distributed-scheme>
<!--
Backing map scheme definition used by all the caches that do
not require any eviction policies
-->
<local-scheme>
<scheme-name>unlimited-backing-map</scheme-name>
</local-scheme>
</caching-schemes>
</cache-config>

Answer 1

user10307225 wrote:
Hi Aleks,
Thanks for your reply.
Is there any way to configure Coherence to do the replication Asynchronously to avoid the replication delays?Hi Adsen,
that is a very bad idea, as that means that your code can return earlier than your data is safe. Now if the primary node died after your code returned but before the data was backed up, your data would be lost.
The main strong point of Coherence is that it safeguards your data, and what you propose makes that theoretically impossible.
If you have concerns about performance then Coherence has multitude of options allowing you to tune the system, but until those performance issues manifest, don't worry about them. And of course stress test your system frequently during development to ensure that if the problems are valid then they do manifest :-)
Also the usual way of improving throughput in high performance systems is batching.
For improving latency, you could
- try to minimize the traffic you need, e.g. send an entry-processor doing the modification instead of sending your entire cached value over the network
- try to have your synchronous code minimal but still retain data safety but e.g. execute the logic asynchronously (with data safety ensuring that it will be executed), e.g. send a command, which will be enqueued synchronously but executed asynchronously (the command pattern can be found in the Coherence Incubator: http://coherence.oracle.com/display/INCUBATOR/Command+Pattern )
Best regards,
Robert