Bug 761805 (GLUSTER-73)

Summary: write-behind does not set a default cache size
Product: [Community] GlusterFS Reporter: Vikas Gorur <vikas>
Component: write-behindAssignee: Raghavendra G <raghavendra>
Status: CLOSED CURRENTRELEASE QA Contact:
Severity: medium Docs Contact:
Priority: low    
Version: mainlineCC: anush, avati, gluster-bugs
Target Milestone: ---   
Target Release: ---   
Hardware: All   
OS: Linux   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: Type: ---
Regression: RTNR Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Attachments:
Description Flags
Effect of write-behind cache-size on throughput none

Description Vikas Gorur 2009-06-25 09:44:37 UTC
The default cache-size in write-behind is 0, which makes it useless. The default should be set to a sensible value. 1MB might be a good value.

Comment 1 Anand Avati 2009-06-30 17:46:25 UTC
> The default cache-size in write-behind is 0, which makes it useless. The default should be set to a sensible value. 1MB might be a good value.

Is the above line a conclusion made from experimental observation, or just an opinion? The behavior of write-behind with 0 cache-size is to "write behind" one request at a time, and keep the following request blocked (not unwound) till the first request reply (nevertheless instantly stack wind the second request towards the server as soon as it arrives) -- not really useless.

Both patches 653 and 655 seem to be separately submitted towards this same bug. Is there a more convincing reason for these patches?

Avati

Comment 2 Vikas Gorur 2009-07-02 03:37:45 UTC
(In reply to comment #1)
> > The default cache-size in write-behind is 0, which makes it useless. The default should be set to a sensible value. 1MB might be a good value.
> 
> Is the above line a conclusion made from experimental observation, or just an
> opinion? The behavior of write-behind with 0 cache-size is to "write behind"
> one request at a time, and keep the following request blocked (not unwound)
> till the first request reply (nevertheless instantly stack wind the second
> request towards the server as soon as it arrives) -- not really useless.

Not an experimental observation, just an opinion.
 
> Both patches 653 and 655 seem to be separately submitted towards this same bug.
> Is there a more convincing reason for these patches?

I've marked my patch as Superseded. Please apply Raghu's patch.

Comment 3 Vikas Gorur 2009-07-06 10:20:48 UTC
Created attachment 34 [details]
This is another attempt

Comment 4 Vikas Gorur 2009-07-06 10:21:41 UTC
As the graph above clearly shows, it is better to set the default cache size to a higher value.

Comment 5 Raghavendra G 2009-07-08 03:28:08 UTC
fix committed in bde0032cfbe51383c697247831e26394ca6cd913.