Hello OrientDB Team,

Just a quick observation. Hazelcast replication is very slow to a point 
where I don't know how one can recommend it for write-heavy applications. I 
had a two node cluster running OrientDB 2.0.8 Community in Azure for about 
12 hours writing over 1.3 million vertices an hour. This was not test data 
that was being written and was real user data with edges and vertices. The 
machines selected for this purpose have 4 cores and 28GB of RAM with 1TB 
SSD disk attached to each of them. The machines were in the same region 
with <1ms latency between machines. OrientDB was started with 6GB of max 
heap and 16 GB of disk cache buffer. After about 11 hours, node 1 was about 
1.2 million vertices behind node 2. Both OrientDB nodes were reporting more 
than sufficient headroom for operation. Here is the default distributed db 
configuration json file used:

{
    "autoDeploy": true,
    "hotAlignment": false,
    "executionMode": "asynchronous",
    "resyncEvery": 15,
    "readQuorum": 1,
    "writeQuorum": 2,
    "failureAvailableNodesLessQuorum": false,
    "readYourWrites": true,
    "clusters": {
        "internal": {
        },
        "index": {
        },
        "*": {
            "servers" : [ "<NEW_NODE>" ]
        }
    }
}

If I set executionMode to "synchronous", one node goes out of sync within 
minutes and the entire setup falls apart. Asynchronous execution mode is 
the only way I could get the sustained writes to succeed without any 
errors. Unless I missed an important configuration, I don't see how it can 
be used in a multi-node master-master cluster with the expectation that all 
nodes will be in sync within a few seconds. I've tried it with 2 and 4 
nodes with various configurations to no avail. In standalone mode, OrientDB 
is fantastic with screaming fast writes and reads. At this point, scaling 
up is about the most practical option with replication handled by 
individual applications on a use case by use case basis.

I welcome the opportunity to be proved wrong at this point. You guys have a 
great product with replication being the most glaring weak point for 
write-heavy applications.

Amir.

-- 

--- 
You received this message because you are subscribed to the Google Groups 
"OrientDB" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to