(Or is it the other way around?)
I did some tests to qualify the hypothesis of Orly Andico about ReiserFS
being much faster than XFS. In particular I do not see any reason why a
choice between the two should matter on testing a single 100Mbps link.
Hardware:
o Pentium III 733MHz
o 3ware Escalade 6400 hardware RAID 5
o 4 x IBM 7200RPM UDMA/66 30GB HDD
o 512MB RAM
Software:
o ProFTPd 1.2.1
o lftp 2.3.11
o Linux kernel 2.4.7-xfs
Legend / Codes:
XFS1 - existing XFS partition used for main bulk of data
XFS3 - existing XFS partition used for home directories
Reiser - new ReiserFS filesystem
XFS2 - new XFS filesystem created in the same place as Reiser
Procedure:
a. mkreiserfs -f -h r5 -v2 /dev/sda7
b. mount /dev/sda7 /opt/test -t reiserfs -o rw,noatime,nodiratime,notail
c. transfer XFS1 -> Reiser
d. transfer XFS1 -> XFS3
e. transfer Reiser -> XFS1
f. transfer XFS1 -> XFS3
g. transfer Reiser -> Reiser
h. mkfs.xfs -f -l size=32768b /dev/sda7
i. mount /dev/sda7 /opt/test -t xfs -o
rw,noatime,nodiratime,osyncisdsync,logbufs=8,biosize=16
i. transfer XFS1 -> XFS2
j. transfer XFS1 -> XFS3
k. transfer XFS2 -> XFS1
l. transfer XFS1 -> XFS3
m. transfer XFS2 -> XFS2
Notes:
o the system was not rebooted before the test was done to simulate a
"real world situation". Unfortunately this may have caused a biase against
ReiserFS during the test (b) because it may have had the need to swap out
a number of applications to make space for its cache. This scenario is not
unlikely, though. In particular no applications were "active", although
standard programs of my server (PostgreSQL, Postfix) were loaded.
o I created a new filesystem for ReiserFS and XFS testing purposes
despite the fact this is not "realistic" because we scrap out the
fragmentation issues. This was done to allow for fairness because both
XFS2 and Reiser were on /dev/sda7 (para patas).
o transfers were done using lftp connecting to localhost. All filesystems
reside on the same RAID5 system. This causes double-traffic because the
system has to read and write. Numbers would probably have been higher had
I had hard drives on separate controllers, doing drive to drive transfers.
But that is beside the point, as we are not testing hardware but the
filesystem performance.
o data transferred was always a unique ISO to prevent the situation were
the file was already in cache. They were similar in size, though. I picked
ISOs that were ~630MB.
o while omitted in the procedure listing, files created during the
transfer, except transfers TO the XFS2 and Reiser filesystems were deleted
after the transfer.
Results:
o XFS1 --> Reiser = 3.88 M/s
o XFS1 --> XFS3 = 5.17 M/s
o Reiser --> XFS1 = 5.88 M/s
o XFS1 --> XFS3 = 5.70 M/s
o Reiser --> Reiser = 4.39 M/s
o XFS1 --> XFS2 = 6.21 M/s
o XFS1 --> XFS3 = 5.35 M/s
o XFS2 --> XFS1 = 6.39 M/s
o XFS1 --> XFS3 = 5.52 M/s
o XFS2 --> XFS2 = 6.56 M/s
Analysis:
o Take note that the redundant tests done to populate the cache with
something other than what the XFS2 and Reiser filesystems will be
subjected to are, while not exact, somewhat close to each other, as they
should be with no significant difference.
o Because Reiser and XFS2 were made on the same partition, they ought to
have the same speed biases as far as hardware location is concerned.
o The microstatistics show significant information to disprove the
hypothesis that ReiserFS is significantly faster than XFS. This does NOT
imply that XFS is faster than ReiserFS. It has enough weight, though, to
show that with files of size 600MB, XFS is conservatively at least at par
with ReiserFS as far as read and write performances are concerned.
--> Jijo
--
Federico Sevilla III :: [EMAIL PROTECTED]
Network Administrator :: The Leather Collection, Inc.
_
Philippine Linux Users Group. Web site and archives at http://plug.linux.org.ph
To leave: send "unsubscribe" in the body to [EMAIL PROTECTED]
To subscribe to the Linux Newbies' List: send "subscribe" in the body to
[EMAIL PROTECTED]
