[SGVLUG] SUSE abandoning ReiserFS

[Mike Fedyk]

Dustin Laurence wrote:
> On Mon, Oct 02, 2006 at 12:55:02PM -0700, Mike Fedyk wrote:
>   
>> The revelation that reiser3 uses the BKL everywhere only makes the 
>> argument stronger.
>>     
>
> Yeah, it's surprising, and given the current CPU arch direction I
> imagine that will get less and less forgivable.
>
> That said, let's be fair: ext2 (but not ext3) still used it a while back
> and may still, presumably since it is also in pretty deep maintenance
> mode.  They might have decided getting rid of the BKL counts as
> maintenance since then, though.  Actually, that's probably the case, it
> has been some time since I knew ext2 still used it.
>   

ext2 and ext3 both had their locking improved during the 2.5 kernel 
development cycle, and it will be improved further in ext4.

>   
>> I have used reiser3 before in the past and liked it until I ran into a 
>> lot of problems with the reiserfsck tools.  The last time I used 
>> reiserfsck, it was anything but robust with a 50% chance of needing to 
>> do a meta-data reconstruction.
>>     
>
> That is the most persistent complaint about reiser.  I may have even
> seen that, since I've seen reiser have recovery trouble but don't recall
> seeing it with ext.  However, unlike apparently the entire rest of the
> planet I actually know my data has no statistical validity and refuse to
> extrapolate from one incident to vast generalizations about Life, The
> Universe, and Everything. :-)
>
> You might have enough data to tell, I'm just poking a stick at all the
> people who say "filesystem X is garbage, use Y, I had one problem with X
> and none with Y."
>
>   

I may have lost ~5 filesystems because of the reiserfsck tools.  That's 
enough for me.

>> I don't think SUSE had a choice.
>>     
>
> It doesn't sound like it.
>
>   
>> bad blocks and and power outages: "don't use xfs unless it is on RAID 
>> and powered by a UPS".  It is an "enterprise" filesystem and it scales 
>>     
>
> Yes, I found that argument compelling.
>
> reiser also has worse latency than ext, which is why I've abandoned
> reiserfs and am back to ext exclusively.
>   

What did you use to measure the latency of the filesystem, and what 
operations were you performing during the measurement?

> Ext2 is still lightning fast for small heavily used filesystems due to
> the lack of journaling.  I like it for /tmp and such, if they're on
> separate filesystems.  I admit I've never benchmarked whether this is
> actually worth it in practice.
Ext2 is on average 1.5 or more times faster than ext3.  ext2 is still 
the performance king to beat when you don't have to worry about what 
happens after a crash or power outage.