Predicting Availability of an In-Memory Computing Cluster

Predicting Availability of an In-Memory Computing Cluster
Project status: finished

Together with the Hasso Plattner Institute in Potsdam as well as with industry partners such as SAP, HP and Fujitsu, in 2014 a project was conducted by the VLBA Lab in Magdeburg in which the availability of an in-memory computing cluster was analyzed. Therefore, a SAP HANA(r) computing cluster was provided by the Future SOC Lab at the HPI.

For this analysis, first the defective components of a single HANA node were modeled in order to estimate the availability of a single HANA system. On that basis, the so-called host auto-failover mechansim was integrated into the model. This mechanism automatically activates passive HANA nodes when one of the active nodes fails. By doing so, availability and performance of the in-memory cluster can be maintained.

Using measured data the model could be parameterized, for instance with takeover time distributions. By simulating the model the relation between the number of active and passive nodes on the one hand as well as the cluster availability on the other hand could be formalized. This formula is intended to support decision makers when planning the introduction of an in-memory computing cluster meeting the availability constraints.