Operate JVM at Scale



OVH has 27 data centers, more than 300 000 servers and 1.3 million customers. We operate one of the world’s largest cloud infrastructures on a daily basis, running millions of applications. Those softwares manage millions of transactions every second. To monitor such an architecture, OVH has chosen to build a unified monitoring platform: OVH Metrics Data Platform. Such a platform uses complex distributed systems to operate, based mainly on open source tools running on the JVM (Apache Kafka, Apache HBase, Warp 10, …). In order to improve developments on such systems, it is useful to have visualization tools to be able to observe the different behaviors of the JVM. This approach makes it possible, for the purpose of continuous improvement and in the interest of performance, to perfect the applications that we develop. This talk summarizes two years of experiments on the development and administration of applications using the JVM in a distributed world.