![SYBASE.csh error on AIX [CR 403816]](arrow-right.gif){kind=icon}
Adaptive Server Enterprise load performance can be affected by both the amount of memory available on the machine and the size of the data set.
ASE must be configured with sufficient memory to hold the complete data set in the data cache. If for some reason ASE is not configured with sufficient memory, writes will go to disk. In such a scenario raw devices should be used for better performance.
With a small data set, if the file system cache is large enough, all data remains in the file system memory and no writes to disk are necessary. With a larger data set, however, the file system cache may become full, causing dirty buffers to be written to disk. The overhead of the writes to disk takes CPU time from both the server and the bcp utility, thus affecting load performance.
In those instances where there is not enough physical memory on the machine, you can improve overall load performance of large data sets by using raw devices, rather than file system devices. Writes are faster while using raw devices, as there is no double buffering as with file system devices. The initial load performance with a raw device may not be as good as with a file system device, but there is no performance drop over time and the total load time is less than the load time with a file system device.
To use raw devices rather than file system devices, change the physname specification in the disk init commands to the names of raw devices. Note that the disk init dsync parameter has no effect when initializing a raw device.
Keep in mind that the volume of data in a data set is considered large or small depending on the amount of memory on your machine. The more memory available to the file system cache, the larger the data set that can be handled efficiently by file system devices. If you can provide adequate memory to the file system cache after configuring ASE with sufficient memory, using raw devices may not be necessary.
| Copyright © 2005. Sybase Inc. All rights reserved. |
|
|
