|
HP OpenVMS Systemsask the wizard |
|
The Question is: How do you determine the 'optimal' cluster size for a large disk volume with a large number of small files to go on it? For example, I have a 44,000,000 block disk (default cluster size of 43), and I'm going to put on it approx 100,000 files of user data. Are there any i/o performance considerations, or anything else, that should worry about if I select a cluster size of 3 (assuming the majority of the 100,000 files will never be greater than 3 blocks)? Finally, while researching this question on the Compaq website (and not finding the answers I was looking for), I came across the following article which implies (to me) that a smaller cluster size was introduced with ODS-5, but I have an ODS-2 volume (un der VMS 7.2/7.2-2) with a lower cluster size (than the default). Is this article in error? The link is: http://www.compaq.com/support/asktima/operating_systems/CHAMP_SRC990406003493.ht ml The Answer is : Optimal for what? If there were such a thing as "Optimal" in isolation, there would be no switches or parameters as they wouldn't be necessary. You can only optimize for a specific operation or set of constraints. When choosing a cluster factor, you might want to optimize for minimal wastage of disk space. The ultimate optimal cluster size is then smaller than the smallest file. If you knew in advance the sizes of all files, you could go one better and choose the best fit common divisor for file sizes, thus minimizing unused space. In terms of I/O, cluster size can have an effect on allocation (large cluster size reduces the amount of work) and on fragmentation, since a file smaller than the cluster size cannot, by definition, be fragmented. So, roughly speaking, larger cluster sizes, more work for the file system, but more wasted space. If you know the approximate size of files in advance, the wizard would suggest you choose a cluster size that just above the mode of the distribution of file sizes. In your case if the majority of files will be under 3 blocks, then choose 3 blocks. A larger cluster factor is unlikely to improve I/O performance, but will waste disk space. Although the DISK may become fragmented into 3 block chunks, for your use that won't matter as FILES won't be fragmented. Small cluster factors were introduced with OpenVMS V7.2, which also introduced ODS-5. ODS-2 disks will, by default choose the historical default for compatibility with pre V7.2 systems, but you can choose a smaller cluster factor by an explicit INITIALIZE/CLUSTER=x. See HELP INIT/CLUSTER on any V7.2 or higher system.
|