xfs
[Top] [All Lists]

Re: How to track down abysmal performance ata - raid1 - crypto - vg/lv -

To: xfs@xxxxxxxxxxx
Subject: Re: How to track down abysmal performance ata - raid1 - crypto - vg/lv - xfs
From: Stan Hoeppner <stan@xxxxxxxxxxxxxxxxx>
Date: Sat, 07 Aug 2010 18:43:29 -0500
In-reply-to: <20100807101349.GD7362@dastard>
References: <20100804085039.GA11671@xxxxxxxxxxxxx> <20100804091317.GA27779@xxxxxxxxxxxxxxxxxx> <20100804092122.GA2998@xxxxxxxxxxxxx> <20100804073546.GA7494@xxxxxxxxxxxxxxxxxxxxxxxxxx> <201008041116.09822@xxxxxx> <20100804102526.GB13766@xxxxxxxxxxxxxxxxxx> <20100804111803.GA32643@xxxxxxxxxxxxx> <alpine.DEB.1.10.1008041351100.19930@xxxxxxxxxxxxxxxx> <20100805082438.0b476adb@notabene> <4C5A774D.6000400@xxxxxxxxxxxxxxxxx> <20100807101349.GD7362@dastard>
User-agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.1.11) Gecko/20100711 Thunderbird/3.0.6
Dave Chinner put forth on 8/7/2010 5:13 AM:
> On Thu, Aug 05, 2010 at 03:33:17AM -0500, Stan Hoeppner wrote:
>> Neil Brown put forth on 8/4/2010 5:24 PM:
>>
>>> Both page-cache and read-ahead work at the filesystem level
>>
>> Are you referring to /sys/block/sdx/queue/read_ahead_kb?  I'm pretty sure 
>> this
>> works below the FS level and below the partition level.  This read_ahead 
>> works
>> at the block device level.  At least for individual or JBOD.
> 
> That number is used to initialise the default readahead value for
> any file descriptor opened on the filesystem. readahead is tracked
> per-fd at the page cache level, so is effectively at the filesystem
> level, not the block device.
> 
>> Are you saying this setting gets ignored by the kernel if/when mdadm, LVM,
>> and/or crypto are used?
> 
> Only the value from the block device the filesystem sits on is used.
> i.e. if you are using /dev/md0, then the filesystem uses the value
> from /sys/block/md0/queue/read_ahead_kb and ignores all the ones set
> on the /dev/sdX devices that make up /dev/md0.

Thanks for the clarification/education Dave.

-- 
Stan

<Prev in Thread] Current Thread [Next in Thread>