Dear all,
unfortunately I observe drastic drops in read- performance when I connect LUNs >2.2T to our Centos 5.5 Servers. I suspect issues with the GUID Partition Table, since it happens reproducibly only on the LUNs >2.2T and goes back to normal performance when using a LUN <2T with "normal", Legacy MBR partitions. All machines are CentOS 5.5 (and RHEL 5.5) on IBM Blades LS21/LS41, all LUNs via Brocade 4G FC Switches from a SUN 7310 Unified Storage, EXT3 filesystems with standard settings.
Any hints?
regards Jens
2010/7/14 Jens Neu jens.neu@biotronik.com:
Dear all,
unfortunately I observe drastic drops in read- performance when I connect LUNs >2.2T to our Centos 5.5 Servers. I suspect issues with the GUID Partition Table, since it happens reproducibly only on the LUNs >2.2T and goes back to normal performance when using a LUN <2T with "normal", Legacy MBR partitions. All machines are CentOS 5.5 (and RHEL 5.5) on IBM Blades LS21/LS41, all LUNs via Brocade 4G FC Switches from a SUN 7310 Unified Storage, EXT3 filesystems with standard settings.
Any hints?
I think ext3 is a bit slow on so big LUNS, try xfs or similar?
-- Eero
I think ext3 is a bit slow on so big LUNS, try xfs or similar?
there is no official support for xfs in RHEL5, so that is not an option. Besides, I'm talking about a drop of >150mb/s to ~10mb/s read performance when cracking the 2.2T size. This is clearly not in the ext3 (or FS for that matter) magnitude of performance issues. Also, numbers stay the same when dd'ing on the raw device (LUN).
-Jens
Eero Volotinen eero.volotinen@iki.fi Sent by: centos-bounces@centos.org 07/14/2010 11:35 AM Please respond to CentOS mailing list centos@centos.org
To CentOS mailing list centos@centos.org cc
Subject Re: [CentOS] GPT Partitions >2.2T with Centos 5.5
Jens Neu writes:
I think ext3 is a bit slow on so big LUNS, try xfs or similar?
there is no official support for xfs in RHEL5, so that is not an option. Besides, I'm talking about a drop of >150mb/s to ~10mb/s read performance when cracking the 2.2T size. This is clearly not in the ext3 (or FS for that matter) magnitude of performance issues. Also, numbers stay the same when dd'ing on the raw device (LUN).
This may not be relevant for you at all, but I have seen a comparable drop in read performance on a new Dell machine (hdparm -t, 25MB/s vs. 130MB/s), completely unrelated to the disk type. In this case, I had to add "pci=nommconf" to the kernel line in grub.conf. This was specific to the RH kernel, live CDs with 2.6.32 or .34 kernels were fine. I also noticed that booting up older IBM Intellistations under CentOS gave a warning about a BIOS bug ("MCFG area at e0000000 is not E820-reserved"; whatever that means :) and not using mmconfig.
Also make sure that all BIOS options are set correctly, if applicable.
--------------------------------------------------------------- This message and any attachments may contain Cypress (or its subsidiaries) confidential information. If it has been received in error, please advise the sender and immediately delete this message. ---------------------------------------------------------------
On Wednesday 14 July 2010, Jens Neu wrote:
I think ext3 is a bit slow on so big LUNS, try xfs or similar?
there is no official support for xfs in RHEL5, so that is not an option. Besides, I'm talking about a drop of >150mb/s to ~10mb/s read performance when cracking the 2.2T size. This is clearly not in the ext3 (or FS for that matter) magnitude of performance issues. Also, numbers stay the same when dd'ing on the raw device (LUN).
Do you see the slow behaviour (10 MiB/s) for all of the device or only for the part that is >2T?
Why use a partition table at all? run dd directly against the device. If this is slow then you have a controller side problem.
/Peter
Do you see the slow behaviour (10 MiB/s) for all of the device or only
for the
part that is >2T?
Why use a partition table at all? run dd directly against the device. If
this
is slow then you have a controller side problem.
very true, back to "Go" :(
Peter Kjellstrom cap@nsc.liu.se Sent by: centos-bounces@centos.org 07/14/2010 01:39 PM Please respond to CentOS mailing list centos@centos.org
To centos@centos.org cc
Subject Re: [CentOS] GPT Partitions >2.2T with Centos 5.5
-
Eero Volotinen -
Jens Neu -
Lars Hecking -
Peter Kjellstrom