Amazon Web Services (AWS) platform can be an unpredictable environment compared to traditional in-house data centers. This is because the machines, databases or CPUs are shared with other unknown applications in the cloud, causing fluctuations. When you gear up your Hazelcast application from a physical environment to Amazon EC2, you should configure it so that any network outage or fluctuation is minimized and its performance is maximized. This section provides notes on improving the performance of Hazelcast on AWS.
Hazelcast is an in-memory data grid that distributes the data and computation to the nodes that are connected with a network, making Hazelcast very sensitive to the network. Not all EC2 Instance types are the same in terms of the network performance. It is recommended that you choose instances that have 10 Gigabit or High network performance for Hazelcast deployments. Please see the below table for the recommended instances.
|Instance Type||Network Performance|
Since data is sent and received very frequently in Hazelcast applications, latency in the network becomes a crucial issue. In terms of the latency, AWS cloud performance is not the same for each region. There are vast differences in the speed and optimization from region to region.
When you do not pay attention to AWS regions, Hazelcast applications may run tens or even hundreds of times slower than necessary. The following notes are potential workarounds.
AWS uses two virtualization types to launch the EC2 instances: Para-Virtualization (PV) and Hardware-assisted Virtual Machine (HVM). According to the tests we performed, HVM provided up to three times higher throughput than PV. Therefore, we recommend you use HVM when you run Hazelcast on EC2.