[Date Prev][Date Next][Thread Prev][Thread Next] [Date Index] [Thread Index]

Subject: ffs, ext2, metadata operations, FreeBSD
From: Christian Jacken
Date: 2 Mar 2000 08:00:00 -0000


Hallo,

soeben las ich:

-----------------------------------------

When the system suddenly, unexpectedly can't write out its data (power
outage, broken harddisk, pilot error etc.), you will obviously lose data
from files your applications just wrote to, due to the write buffering any
modern operating system does.
Now, that's only part of the story. Besides the data inside your files,
there are also so-called "metadata" operations, i.e. creating, deleting and
renaming files.

These operations are much more dangerous when they're interrupted. We're not
just talking losing data from those files you just wrote, we're talking
losing files you didn't even touch in last days! The "meta-data" operations
can cause you to lose your whole filesystem (partition) or they can delete
or truncate files that just happened to be "neighbors" of the files you
wrote.

Hence, operating system usually apply different write buffer mechanisms for
metadata than for normal data (the data your applications write into their
files).

If you are too conservative (i.e. if you write out all metadata immediatly)
the systems is very slow for some operations, for example unpacking files
that have many files inside them (zip or tar files). All BSD-derived OSes
and OSes with the BSD filesystem default to an almost- immediatly mode
(called "synchronous"), which makes them very slow for those operations.

In Linux, metadata is written out as asynchronous/delays as normal is, which
is the other extreme in the saveness/speed tradeoff.

------------------------------------------

Meine Fragen: Wie kann ich unter Linux metadata writing eine sicherere
Einstellung wählen?

Kann jemand sich zu Performanceunterschieden zwischen ext2, reiserfs und ffs
(FreeBSD native fs) äußern?

Vielen Dank,

Christian

P.S. BTW, nach gründlicher Analyse haben wir uns für FreeBSD als Webserver
entschieden. Mal sehen, wie das in der Praxis klappt. Subjektiv erscheint
FreeBSD gegenüber RH spürbar schneller, was an dem kleineren Code liegen mag
(reduziert auch die Wahrscheinlichkeit von Bugs).

P.P.S. Das Abstimmungsergebnis zu Postings zu Microsoft lautete 4 (Gegner)
gegen 2 (Befürworter). Damit werde ich also die Weiterleitung solcher
Nachrichten einstellen.