Project

General

Profile

Actions
Copy link

Support #64442

open

Ceph stripe parallel write

Added by Nishit Khosla 3 months ago. Updated 2 months ago.

Status:
New
Priority:
Normal
Assignee:
Venky Shankar
Category:
-
Target version:
-
% Done:

0%

Tags:
Reviewed:
Affected Versions:
Ceph - v17.2.5
Component(FS):
libcephfs
Labels (FS):
Pull request ID:

Description

Hello,

We are trying to do performance troubleshooting for cephfs and exploring the option to use stripping. We have set the stripe_count of the cephfs to 4 however we are unable to ascertain if the writes are going to the objects in parallel.

We observed that the performance with stripe_count=1 is same as performance of stripe_count=4. We could observe that with either settings, the IO is happening on all OSDs.

We would like to know how to be sure that writes are happening in parallel on the OSD.

Actions
Copy link
 #1

Updated by Venky Shankar 2 months ago

  • Assignee set to Venky Shankar
Actions
Copy link
 #2

Updated by Venky Shankar 2 months ago

Hi Nishit,

Nishit Khosla wrote:

Hello,

We are trying to do performance troubleshooting for cephfs and exploring the option to use stripping. We have set the stripe_count of the cephfs to 4 however we are unable to ascertain if the writes are going to the objects in parallel.

We observed that the performance with stripe_count=1 is same as performance of stripe_count=4. We could observe that with either settings, the IO is happening on all OSDs.

This support tracker lacks many details.

Mind sharing what tool is being used to assess the IO throughput? We would also need details like max_mds setting, number of OSDs, network bandwidth, user-space or kclient, number os clients to start with a few.

Actions
Copy link

Also available in: Atom PDF