[9fans] devfs

[9fans] devfs

[Christopher Nielsen]

>From reading the code, it looks like mirrors in devfs
are read preferring the first drive in the mirror, and
only reading from the second drive if there's a
failure. Is this correct, and if so, is anyone working
on round-robin reading to improve read performance?

-- 
Christopher Nielsen
"They who can give up essential liberty for temporary
safety, deserve neither liberty nor safety." --Benjamin Franklin

Re: [9fans] devfs

[andrey mirtchovski]

>>From reading the code, it looks like mirrors in devfs
> are read preferring the first drive in the mirror, and
> only reading from the second drive if there's a
> failure. Is this correct, and if so, is anyone working
> on round-robin reading to improve read performance?
> 

i believe that is true. 

would reading every other block from the second one improve
performance?  seems to me it would have the opposite effect, causing
many more seeks than normal (i.e.  there's no hd cache optimization
for 'read every odd/even block from the platter').

i see one clear benefit though, and that is if you have two or more
bulk readers and serve each one from a separate disk.

i'd like to hear other's opinions on this one :)

andrey

Re: [9fans] devfs

[Eric Van Hensbergen]

On Mon, 22 Nov 2004 21:30:01 -0700, andrey mirtchovski
<mirtchov@cpsc.ucalgary.ca> wrote:
> >>From reading the code, it looks like mirrors in devfs
> > are read preferring the first drive in the mirror, and
> > only reading from the second drive if there's a
> > failure. Is this correct, and if so, is anyone working
> > on round-robin reading to improve read performance?
> >
> 
> i believe that is true.
> 
> would reading every other block from the second one improve
> performance?  
> 

What does RAID-1 do?  Maybe every other block isn't the right
granularity, but it seems like farming out requests between the
multiple drives would be the right approach.

         -eric

Re: [9fans] devfs

[Fco. J. Ballesteros]

[-- Attachment #1: Type: text/plain, Size: 291 bytes --]

In our case, we setup mirrors for different partitions in the opposite
order.  For example, fossil is sdC0 then sdD0, but venti is sdD0 then
sdC0.  This seems to be enough to keep both disks being used at the
same time.

Anyway, nobody here has ever tried to optimize disk usage here.

[-- Attachment #2: Type: message/rfc822, Size: 3616 bytes --]

From: Eric Van Hensbergen <ericvh@gmail.com>
To: Fans of the OS Plan 9 from Bell Labs <9fans@cse.psu.edu>
Subject: Re: [9fans] devfs
Date: Tue, 23 Nov 2004 07:53:36 -0600
Message-ID: <a4e6962a041123055326576945@mail.gmail.com>

On Mon, 22 Nov 2004 21:30:01 -0700, andrey mirtchovski
<mirtchov@cpsc.ucalgary.ca> wrote:
> >>From reading the code, it looks like mirrors in devfs
> > are read preferring the first drive in the mirror, and
> > only reading from the second drive if there's a
> > failure. Is this correct, and if so, is anyone working
> > on round-robin reading to improve read performance?
> >
> 
> i believe that is true.
> 
> would reading every other block from the second one improve
> performance?  
> 

What does RAID-1 do?  Maybe every other block isn't the right
granularity, but it seems like farming out requests between the
multiple drives would be the right approach.

         - eric

[9fans] Re: devfs

[Dave Eckhardt]

> From reading the code, it looks like mirrors in devfs
> are read preferring the first drive in the mirror, and
> only reading from the second drive if there's a
> failure. Is this correct, and if so, is anyone working
> on round-robin reading to improve read performance?

Beware the text beginning with "Since writes go to all
devices in the mirror but reads are performed just by
the first device ..." in the "Setting Up Fossil" Wiki
document,
<http://cm.bell-labs.com/wiki/plan9/setting_up_fossil/index.html>.

Dave Eckhardt