The CPU frequency is practically not growing, but everything else: network/ssd/ data volumes are growing exponentially. This poses new challenges for the databases: reduce CPU costs for working with the infrastructure. The bottleneck of MPP systems is the network. What happens if the network in Greenplum becomes 10 times faster? We accelerated 5 times. Then we tried DPDK, but realized that we couldn’t move forward without serious core refinement. I would share what happened to the database when the network accelerated 5 times. And also why DPDK didn’t start. What conclusions have we learned and what are we planning to do next (spoiler: all new changes will be in Apache Cloudberry).

Speakers:

Leonid Borchuk: Developer at Yandex Cloud, Moscow. Apache Cloudberry commiter

Team lead of the MPP postgres development team at Yandex.Cloud. He started his career with the administration of large commercial universal and analytical databases. But now he’s finally writing his own. Apache Cloudberry commiter.