bugGluster - Bugs: bug #17280, XLATOR - getattr prefetch

 
 

bug #17280: XLATOR - getattr prefetch

Submitted by:  Anand Avati <avati>
Submitted on:  Thu 03 Aug 2006 02:47:23 PM UTC  
 
Category: GlusterFSSeverity: 4 - Important
Priority: 5 - NormalItem Group: Improper behaviour
Status: FixedPrivacy: Public
Assigned to: Basavanagowda Kanur <gowda>Originator Name: 
Open/Closed: ClosedRelease: 
Operating System: GNU/LinuxReproducibility: None

Add a New Comment(Rich Markup)
   

You are not logged in

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

 

Thu 14 Sep 2006 01:59:15 PM UTC, comment #5:

NOTE - flush the cache held whenever any entry in the cache undergoes a change (chmod, unlink, chown) or if a new file is created in the directory (open, mknod, symlink, link, rename)

Anand Avati <avati>
Project Member
Thu 14 Sep 2006 01:57:30 PM UTC, comment #4:

publish performance figures with and without stat-prefetch translator for the combinations of cases -

1. large timeout for cache
2. small timeout for cache
3. directories with large entries
4. directories with small number of entries
5. parallel ls -R

also inspect if performance can be improved by internal restructuring of data structures.

check for thread safety.

Anand Avati <avati>
Project Member
Sun 06 Aug 2006 10:47:33 PM UTC, comment #3:
  • getattr xlator implementation with prefetching using getattr for individual entries in directory is not giving any performance improvement as the number of network calls is still the same.
Basavanagowda Kanur <gowda>
Project MemberIn charge of this item.
Sat 05 Aug 2006 10:32:40 AM UTC, comment #2:

to confirm on the previous comment, i observed that network throughput during the 'ls -l' is just 60kbit/sec. Clearly there is a need to perform 'bulk' operation as said before to exploit network i/o.

Anand Avati <avati>
Project Member
Sat 05 Aug 2006 10:29:20 AM UTC, comment #1:

the current getattr xlator is not giving the expected performance boost.

i observed that multiple getattr calls over the network is actually slowing the process.

need to enhance readdir to embed each entry's stat along with the entry and that should be cached for future getattr calls. Calling multiple getattr within readdir is not solving. One network transaction should get all entries and their stat() in one shot.

Anand Avati <avati>
Project Member
Thu 03 Aug 2006 02:47:23 PM UTC, original submission:

getattr prefetching should be done preferably when a readdir() is done expecting that getattr will be performed on those entries next

Anand Avati <avati>
Project Member

 

(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 avati (Updated 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 10 latest changes.

    Date Changed By Updated Field Previous Value => Replaced By
    Sun 06 May 2007 10:43:19 AM UTCavatiStatusIn Progress=>Fixed
      Open/ClosedOpen=>Closed
    Sat 05 Aug 2006 10:32:40 AM UTCavatiStatusFixed=>In Progress
    Sat 05 Aug 2006 10:29:20 AM UTCavatiItem GroupMissing Feature=>Improper behaviour
      Open/ClosedClosed=>Open
    Fri 04 Aug 2006 06:57:03 PM UTCgowdaStatusIn Progress=>Fixed
      Open/ClosedOpen=>Closed
    Thu 03 Aug 2006 07:48:57 PM UTCavatiSeverity3 - Normal=>4 - Important
      StatusNone=>In Progress
      Assigned toNone=>gowda

    Back to the top


    Powered by Savane 3.1-cleanup1