bugGluster - Bugs: bug #25186, afr reads and writes in 4kb chunks

 
 

bug #25186: afr reads and writes in 4kb chunks

Submitted by:  swankier <swankier>
Submitted on:  Sat 27 Dec 2008 09:00:10 PM UTC  
 
Category: GlusterFSSeverity: 3 - Normal
Priority: 5 - NormalItem Group: Missing Feature
Status: ConfirmedPrivacy: Public
Assigned to: NoneOriginator Name: swankier
Open/Closed: ClosedRelease: 
Operating System: GNU/LinuxReproducibility: Every Time

Add a New Comment(Rich Markup)
   

You are not logged in

Please log in, so followups can be emailed to you.

 

(Jump to the original submission Jump to the original submission)

Tue 03 Feb 2009 12:04:18 AM UTC, comment #7:

increased to 128KB for faster self-heal. will be done as per the other factors in future releases.

Amar Tumballi <amarts>
Project Member
Mon 05 Jan 2009 08:48:31 AM UTC, comment #6:

this value should be tied to the tcp window value if at all possible to reduce the amount of tcp overhead.

The amount of tcp chatter with gluster is quite high. I'm trying to use gluster over a WAN link with high bandwidth but high latency (a vpn.) This extra chatter has a pretty large impact on file throughput (especially for many small files.)

swankier <swankier>
Mon 05 Jan 2009 06:46:43 AM UTC, comment #5:

it means that files transfer at incredibly slow speeds.

like... less than one MBps.

swankier <swankier>
Fri 02 Jan 2009 08:38:59 AM UTC, comment #4:

Amar, possibly to have an option that overrides this?

keliew <keliew>
Fri 02 Jan 2009 01:44:52 AM UTC, comment #3:

Hi Swankier,
This has been done intentionally, as we want to bring in support for healing of 'sparse' files (or files with holes), so even striped files can be properly healed.
We decided to go ahead with the performance hit caused by this, because 'self-heal' is not a common operation, and while its happening, we thought the performance is not the primary goal but the integrity of data which matters most.
Let me keep this bug report open, till we all come to a conclusion about whats better approach.

Regards,
Amar

Amar Tumballi <amarts>
Project Member
Tue 30 Dec 2008 09:35:55 AM UTC, comment #2:

Suggestion: to have afr self-heal adhering to aggregate-size and page-size from writebehind/readahead.

keliew <keliew>
Tue 30 Dec 2008 04:03:06 AM UTC, comment #1:

This is due to the line:

sh->block_size = sh->buf[source].st_blksize;

from the file:

./xlators/cluster/afr/src/afr-self-heal-data.c

I changed the entry to:

sh->block_size = 131072;

and self-heal became much quicker.

swankier <swankier>
Sat 27 Dec 2008 09:00:10 PM UTC, original submission:

It seems that writeback has little impact on this.

It would be nice for either writeback to have an impact on afr self-heal, or for there to be a tunable to change this size.

swankier <swankier>

 

(Note: upload size limit is set to 16384 kB, after insertion of the required escape characters.)

Attach File(s):
   
   
Comment:
   

No files currently attached

 

Depends on the following items: None found

Items that depend on this one: None found

 

Carbon-Copy List
  • -unavailable- added by amarts (Posted a comment)
  • -unavailable- added by keliew (Posted a comment)
  • -unavailable- added by swankier (Submitted the item)
  •  

    Do you think this task is very important?
    If so, you can click here to add your encouragement to it.
    This task has 0 encouragements so far.

    Only logged-in users can vote.

     

    Please enter the title of George Orwell's famous dystopian book (it's a date):

     

     

    Follow 3 latest changes.

    Date Changed By Updated Field Previous Value => Replaced By
    Tue 03 Feb 2009 12:04:18 AM UTCamartsOpen/ClosedOpen=>Closed
    Fri 02 Jan 2009 01:44:52 AM UTCamartsStatusNone=>Confirmed
      Originator Name=>swankier

    Back to the top


    Powered by Savane 3.1-cleanup1