1 Million CQL operations per server

DevconTLV March 2016 Conference, Tuesday, March 22, 2016, 11:25

CPU core counts continue to grow, along with the raw speed of networking and storage devices available on a modern system. Software design approaches that were valid and safe even a few years ago are no longer sustainable. On new hardware, the performance of standard workloads depends more on locking and coordination across cores than on performance of an individual core.

The team behind the KVM hypervisor and the OSv unikernel is moving up the stack. The Seastar platform enables extreme high-throughput, low-latency applications on Linux, and can make existing real-world workloads run completely asynchronously.

Seastar uses shared-nothing data structures that eliminate costly locking between CPUs, and a dedicated user-space TCP implementation that runs on DPDK. These radical changes in the server design translate into 5X-10x performance gain while preserving all of the Cassandra goodies

Dor is the CEO of ScyllaDB, a Cassandra-compatible NoSQL data store that can handle 1 million transactions per second on a single serve. As a young CEO (is 41 young?) I do it all, coding, reviewing, marketing, hiring, testing and even the dishes. Prior to ScyllaDB, Dor was on board the KVM dev team from birth till the glory days. Dor managed the KVM and Xen development teams at Red Hat. Prior to the virtualization era, Dor amused himself by building blade centers solutions and terabit routers.

Other Presentations at DevconTLV March 2016

Open Accessibilty Menu