Performance of a database manager in a virtual memory system

Abstract

Buffer space is created and managed in database systems in order to reduce accesses to the I/O devices for database information. In systems using virtual memory any increase in the buffer space may be accompanied by an increase in paging. The effects of these factors on system performance are quantified where system performance is a function of page faults and database accesses to I/O devices. This phenomenon is examined through the analysis of empirical data gathered in a multifactor experiment. The factors considered are memory size, size of buffer space, memory replacement algorithm, and buffer management algorithm. The improvement of system performance through an increase in the size of the buffer space is demonstrated. It is also shown that for certain values of the other factors an increase in the size of the buffer space can cause performance to deteriorate.

An extension of the performance of a database manager in a virtual memory system using partially locked virtual buffers

“…Total I/O was defined in [4] as the sum of database faults and page faults in the system. The number of database faults is dependent on the virtual buffer size and the buffer management algorithm.…”

“…Reference paging refers to page faults caused by attempts to use information that is in the virtual buffer but not in real memory. The double paging rate is defined in [4] to be the number of double page faults divided by the number of database faults. A similar definition for the reference paging rate was also given in [4].…”

“…The PRIME 300 has the same configuration and uses the same software probe in both sets of experiments. A detailed description of the experimental environment can be found in [4],…”

VMMB: Virtual Machine Memory Balancing for Unmodified Operating Systems

“…In VMM-level swapping, the VMM performs page replacement on the guest physical memory. However there could be a pathological performance degradation known as the double paging anomaly [21,22]. When the guest tries to swap out a page that is already swapped out by the VMM, the page needed to be swapped in by the VMM.…”

“…Finally, we use a combination of ballooning [8] and VMM-level swapping in order to efficiently select the victim pages and to immediately allocate memory to a beneficiary VM. For an efficient VMM-level swapping, we introduce a technique to avoid double paging anomaly [21,22] that can seriously degrade the performance when the same page is selected as a victim by the VMM and VM.…”