A Queueing Network Model for Performance Prediction of Apache Cassandra

Abstract

NoSQL databases such as Apache Cassandra have attracted large interest in recent years thanks to their high availability, scalability, flexibility and low latency. Still there is limited research work on performance engineering methods for NoSQL databases, which yet are needed since these systems are highly distributed and thus can incur significant cost/performance trade-offs. To address this need, we propose a novel queueing network model for the Cassandra NoSQL database aimed at supporting resource provisioning. The model defines explicitly key configuration parameters of Cassandra such as consistency levels and replication factor, allowing engineers to compare alternative system setups.Experimental results based on the YCSB benchmark indicate that, with a small amount of training for the estimation of its input parameters, the proposed model achieves good predictive accuracy across different loads and consistency levels. The average performance errors of the model compared to the real results are between 6% and 10%. We also demonstrate the applicability of our model to other NoSQL databases and other possible utilisation of it.

PAX: Partition-aware autoscaling for the Cassandra NoSQL database

“…We have observed that the demand D changes almost linearly with the number of active nodes, as shown in Figure 5. This is mainly due to the increasing number of read requests issued by the request coordinator [28]. In our tests, each additional active node increases D on average by α = 10%.…”

Performance Modelling and Optimisation of NoSQL Database Systems

“…For example, a job that visits multiple times the same node, requesting different execution requirements at each visit, may be modelled by assuming that the job switches class in-between visits. Distributed NoSQL databases such as Apache Cassandra provide an example, in which class-switching can be used to express a workflow of execution through multiple nodes in order to retrieve the data needed by a query for its completion [DCS17].…”

SD: A Divergence-Based Estimation Method for Service Demands in Cloud Systems

“…The Jensen Shannon divergence is bounded between 0 ≤ D JS ≤ 1 In this section, we illustrate the Cassandra model used for evaluating the SD algorithm. This is a simplified version of the model presented in [5]. With the aim to reduce the model complexity and the system state space, we have developed a model able to support only the Consistency Level ONE.…”

“…Each Cassandra node uses the processor sharing (PS) scheduling policy. All the other stations used in [5], such as the networks and disk queues, have been grouped in a single infinite queue (or infinite server), called 'Net' positioned right after the workload generator. In addition, the workload generator has also been modelled as an infinite server.…”

“…For example, a job that visits multiple times the same node, requesting different execution requirements at each visit, may be modelled by assuming that the job switches class in-between visits. Distributed NoSQL databases such as Apache Cassandra provide an example, in which class-switching can be used to express a workflow of execution through multiple nodes in order to retrieve the data needed by a query for its completion [5].…”