Accelerating Analytical Processing in MVCC using Fine-Granular\n High-Frequency Virtual Snapshotting

Efficient transactional management is a delicate task. As systems face\ntransactions of inherently different types, ranging from point updates to long\nrunning analytical computations, it is hard to satisfy their individual\nrequirements with a single processing component. Unfortunately, most systems\nnowadays rely on such a single component that implements its parallelism using\nmulti-version concurrency control (MVCC). While MVCC parallelizes short-running\nOLTP transactions very well, it struggles in the presence of mixed workloads\ncontaining long-running scan-centric OLAP queries, as scans have to work their\nway through large amounts of versioned data. To overcome this problem, we\npropose a system, which reintroduces the concept of heterogeneous transaction\nprocessing: OLAP transactions are outsourced to run on separate (virtual)\nsnapshots while OLTP transactions run on the most recent representation of the\ndatabase. Inside both components, MVCC ensures a high degree of concurrency.\nThe biggest challenge of such a heterogeneous approach is to generate the\nsnapshots at a high frequency. Previous approaches heavily suffered from the\ntremendous cost of snapshot creation. In our system, we overcome the\nrestrictions of the OS by introducing a custom system call vm_snapshot, that is\nhand-tailored to our precise needs: it allows fine-granular snapshot creation\nat very high frequencies, rendering the snapshot creation phase orders of\nmagnitudes faster than state-of-the-art approaches. Our experimental evaluation\non a heterogeneous workload based on TPC-H transactions and handcrafted OLTP\ntransactions shows that our system enables significantly higher analytical\ntransaction throughputs on mixed workloads than homogeneous approaches. In this\nsense, we introduce a system that accelerates Analytical processing by\nintroducing custom Kernel functionalities: AnKerDB.\n