Adaptive Server large memory support on 32-bit Enterprise Linux operating systems increases the amount of available memory in Adaptive Server from 2.7GB to as much as 64GB. Increasing the amount of memory available to Adaptive Server improves performance by significantly reducing the number of times the server must access the disk.
Adaptive Server large memory support makes use of the shared memory file system (shmfs) and the memory mapped file (mmap) features available on Linux 32-bit operating systems. When large memory support is enabled, Adaptive Server creates an shmfs file of a size specified in the Adaptive Server configuration file. This file can be up to 16GB for Red Hat Advanced Server Linux 2.1 (AS 2.1) operating systems and up to 64GB for Red Hat Enterprise Linux 3.0 (RHEL 3) operating systems.
Adaptive Server uses memory-mapped file space for a secondary data cache. The primary data cache can only be as large as the 2.7GB, which is the maximum addressable memory allowed by Linux, minus any necessary overhead.
Adaptive Server always places new pages in the primary data cache. When the primary cache is full, Adaptive Server relocates a page to the secondary cache to make room for the new page. Adaptive Server searches for a given page first in the primary data cache and then in the secondary cache. If the page is found in the secondary cache, Adaptive Server maps a virtual address window to that portion of the shmfs, and copies the page to the primary cache where it can be read.
The secondary cache:
Is global; it is shared by all named primary caches.
Can replace only pages of the same size as the server page size. Thus, if the server page size is 4K and there are two pools, one with a page size of 4K and the other with a page size of 16K, the secondary cache holds only pages from the 4K pool.
Supports writes only for washing and checkpointing.
Has a fixed wash size of 20%. Wash size for pools in primary caches are not honored—all washes take place in the secondary cache.
Does not support index trips and OAM trip tuning.
Will not hold pages from caches marked “log only” or “relaxed lru” because such caches do not benefit from the secondary cache. Log pages are typically not read back from disk, and caches are marked “relaxed lru” only when the object is fully cached in the primary cache.
| Copyright © 2004. Sybase Inc. All rights reserved. |
|
|
