[Top] [All Lists]

Re: File system remain unresponsive until the system is rebooted.

To: xfs@xxxxxxxxxxx
Subject: Re: File system remain unresponsive until the system is rebooted.
From: Michael Monnerie <michael.monnerie@xxxxxxxxxxxxxxxxxxx>
Date: Wed, 15 Feb 2012 12:38:52 +0100
Cc: Peter Grandi <pg_xf2@xxxxxxxxxxxxxxxxxx>
In-reply-to: <20267.5137.85650.499331@xxxxxxxxxxxxxxxxxx>
Organization: it-management http://it-management.at
References: <CANs4eSBWLc4HxAbPZ8kOVOdJ7RKiA+-ai3Q2J+FAyuzHtUqfdg@xxxxxxxxxxxxxx> <20265.9379.139218.148520@xxxxxxxxxxxxxxxxxx> <20267.5137.85650.499331@xxxxxxxxxxxxxxxxxx>
User-agent: KMail/4.7.2 (Linux/3.1.9-1.4-desktop; KDE/4.7.2; x86_64; ; )
Am Donnerstag, 2. Februar 2012, 22:54:09 schrieb Peter Grandi:
> This then the argument that on platforms with bad latency that
> decision works still works well because then you might as well go
> for throughput.

Hi, I just took these lines to reply to your whole mail. I guess that
the advantage of XFS will grow on a shared storage type like you
typically have on a VM environment. The aggregation XFS does can result
in a more bursty type of I/O, with larger I/Os happening at once. That
always is better for RAID storage - which you normally have in a VM
environment. Also, all better RAID controllers, and especially
enterprise RAIDs, have large write buffers, so even more aggregation
occurs at the storage itself, helping throughput maximisation.

I don't know of any scientific investigation of "which filesystem is
better in a VM environment" that could be referenced in a generic way,
mostly because there are so many variables there that it doesn't
neccessarily fit your own use case. Maybe someone can point me to such
research material.
My hope is - and that is what Dave is arguing - that minimising I/O
"disturbances" by metadata work like log file handling helps keeping
overall throughput on a shared storage type in a VM environment high.
And that seems very reasonable.
I don't really understand your argument about delay for a single thread
fsync. First, XFS should do this quicker by "batching" transactions, and
second, overall storage throughput is usually much more important than
that of a single server performance - at least in a VM environment. I
need to run 50 servers on a storage with acceptable performance, and if
one server needs more performance than is available, you need to do
something else - there are lots of options then.

mit freundlichen Grüssen,
Michael Monnerie, Ing. BSc

it-management Internet Services: Protéger
http://proteger.at [gesprochen: Prot-e-schee]
Tel: +43 660 / 415 6531

Attachment: signature.asc
Description: This is a digitally signed message part.

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