Are you interested in helping with a project rewriting the PCFS driver ?
In short, PCFS for FAT32 s*cks. It's nowhere close to the performance it could have. Now FAT32 surely has abysmal (non-)performance in the degraded case (think a 4GB file made from 8 million 512-Byte fragments), but it needn't be _that_ bad for "usual" fragmentation levels.
No, there's nothing you can tune. Your numbers are pretty high for PCFS already ...
FrankH. On Wed, 8 Nov 2006, Mike Pleasance wrote:
I'm working on a project that uses a fibre-channel raid across several systems (not simultaneously). The raid is formatted with a FAT32 file system so that it can be accessed from VxWorks, Windows and Solaris. We are running Solaris 10 06/06 for Sparc. The raid is capable of a raw data rate in excess of 100 MB/s, and using a windows PC, I can read and write FAT32 data around 70 MB/s. However, on Solaris I can not get anywhere near the performance we need. Here are the numbers I have been getting so far: Writing 500 1MB Files: 4.5 MB/s (majority of the time is spent creating files)
Concurrent creates ?
Writing 1 500MB File: 23 MB/s Reading 500 1MB Files: 26 MB/s
Again, concurrent or in sequence ?
Reading 1 500MB File: 30 MB/s Are there any known performance issues with using FAT32 or are any tricks that could increase the performance? I have tried using non-buffered I/O (using setvbuf) and memory mapping the files (using mmap). Nothing has had any significant effect on the performance.
More than plenty.
Any help or insight would be greatly appreciated!
As said, it's in dire need of an overhaul ...
Thanks, Mike Pleasance This message posted from opensolaris.org _______________________________________________ opensolaris-discuss mailing list [email protected]
_______________________________________________ opensolaris-discuss mailing list [email protected]
