[Gluster-users] Very slow directory listing and high CPU usage on replicated volume

[Joe Julian]

On 11/06/2012 04:38 AM, Joe Landman wrote:
> On 11/06/2012 04:35 AM, Fernando Frediani (Qube) wrote:
>> Joe,
>>
>> I don't think we have to accept this as this is not acceptable thing.
>
> I understand your unhappyness with it.  But its "free" and you 
> sometimes have to accept what you get for "free".
>
I slightly disagree. Yes, it's free but it's free from an organization 
that's determined to take input from it's users and try to make things 
better for them. Testing, filing bug reports, filing enhancement 
requests, and if you have the skills offering patches: these are the 
responsibility of the users in any community developed software. 
"Bitching"^1 about deficiencies in an email thread or blog post without 
having done those other things is a waste of everyone's bandwidth.
>> I have seen countless people complaining about this problem for a
>> while and seems no improvements have been done. The thing about the
>> ramdisk although might help, looks more a chewing gun. I have seen
>> other distributed filesystems that don't suffer for the same problem,
>> so why Gluster have to ?
>
> This goes to some aspect of the implementation.  FUSE makes metadata 
> ops (and other very small IOs) problematic (as in time consuming).  
> There are no easy fixes for this, without engineering a new kernel 
> subsystem (unlikely) to incorporate Gluster, or redesigning FUSE so 
> this is not an issue.  I am not sure either is likely.
>
> Red Hat may be willing to talk to you about these if you give them 
> money for subscriptions.  They eventually relented on xfs.
>
Besides that, a metadata op is small. Add network overhead and it's much 
slower than local filesystems could ever be. If you want rid of that 
problem, use IB rdma.

I'll grant that the documentation says that rdma is not supported. What 
I've come to understand that meaning is that it's just as unsupported as 
any previous version was. After the acquisition, untested features were 
deemed "technology preview" and were "unsupported" as that. Once they've 
gone through RH QA they can be removed from that status. Apparently rdma 
wasn't high enough on the priority list to be tested before the last RHS 
release, thus that status remains. I don't think it works any less than 
it did with 3.2. I've seen various success with IB and wonder if it's 
hardware related.

^1 I'm not targeting this email thread with this word. For the most part 
this seems to be a productive and thoughtful discussion. It's just a 
general denouncement of this behavior across the internet for all 
community developed software.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20121106/c0248c2f/attachment.html>