UFS Is FreeBSD UFS not good?

ralphbsz

Daemon

Reaction score: 1,166
Messages: 1,879

Many buffer cache implementation have a delay before writing starts. From an overall performance sense, that makes great sense: files are often modified again, read again, or deleted, right after writing. Waiting 10 or 30 seconds has a good chance to eliminate useless work, or delay necessary work to a time when the system is more quiet. And if the application cares, it can always force the write to happen sooner.

Actually, I'm quite surprised that FreeBSD finished the write just as the cp command finished. It could be that this is specific to the FAT file system. But I would not characterize FreeBSD's cache mechanism is general as "write through"; that would lead to ridiculously bad performance. To begin with, on a disk that rotates at 7200 rpm, you on average have to wait 40ms for a write, leading to only about 240 writes per second, and with normal IO sizes, this would be laughably slow.
 

Space_Cadet

New Member


Messages: 1

giahung1997, that would not be the best choice of words to describe how you feel about it, but no one is going to attack you.

I use UFS on my all 5 of my FreeBSD machines and am satisfied with how it performs. I can do a hard reboot and it not hose my filesystem, and I've never had corrupted or lost files.
Hi Trihex thanks for the comment. I use UFS and find it fast and reliable. Except under some circumstances where I am building and linking, and the linker does not link in the new image, it seems to find the old file image even though it has been replaced on disk. Eventually like hours later, it does link it properly. Can you comment? Thanks.
 

olli@

Well-Known Member
Developer

Reaction score: 315
Messages: 375

UFS2 on FreeBSD is rock-stable and has excellent performance. I'm mostly using it with SUJ enabled (soft updates journaling; see the -j flag of newfs and tunefs) that provides reliability and performance at the same time.

It should also be noted that UFS is actively developed and improved. Just recently, Kirk started adding an inode hash / checksum feature.
 
Top