I think your blob file performance may greatly depend upon the file system that it used and the workload.
I found this article: http://oss.sgi.com/projects/xfs/papers/filesystem-perf-tm.pdf Andreas Volz <[EMAIL PROTECTED]> wrote: Am Tue, 13 Nov 2007 07:18:19 -0600 schrieb John Stanton: > In a cache situation I would expect that keeping the binary data in > files would be preferable because you can use far more efficient > mechanisms for loading them into your cache and in particular in > transmitting them downstream. Your DB only needs to store a pathname. > > Just be wary of directory size, and do not put them all in the one > directory. I noticed that problem in my current situation. I don't know the file number and size limit in Linux or Windows, but I'm sure there is a limit. My main problem is to find a good algorithm to name the cached files and split them into directories. My current idea is: 1) Put the URL into DB 2) Use a hash function to create a unique name for the cache file 3) Insert the hash name into the same row as the URL The problem with many files in a directory: 4) Use e.g. 'modulo 11' on the URL hash value to get one of ten directory names where to find a file. But this has the drawback to have a static number of cache directories. The algorithm isn't scalable with growing files. Do you think is a good way? Or do you've another idea? regards Andreas ----------------------------------------------------------------------------- To unsubscribe, send email to [EMAIL PROTECTED] -----------------------------------------------------------------------------
