Hello,
I'm a linux sysadmin working with Linux in a large company, and we are
encountering what seems to be a regression bug with the Megaraid SAS
RAID driver.
The issue is performance related , when we compile the linux kernel in
the i386 (32 bits) architecture and activate the 64Gb High Memory
support. The hardware we are using is the Dell M610 Poweredge blade server.
Here is the exact description of the problem :
- we compile the kernels by typing 'make defconfig' , and then 'make
nconfig' and changing only two things :
- the two drivers we need for the kernel to work on the server (NIC
and 'SCSI low-level drivers' -> 'LSI Logic MegaRAID SAS RAID Module').
- the architecture and the support for high memory .
- when we run the kernel, here are the performances we obtain :
With 3.3.x and 3.4.x kernels (including the latest 3.4.16 kernel)
-----------------------------------------------------------------
- when compiled for the x86_64 architecture, or for the i386 32 bits
architecture with 4Gb high memory support , we obtain some good
performance for disk data transfer ( more or less 250 Mb/s ).
- but with i386 architecture and 64 Gb High Memory support ,
performances decrease in a +200 factor : we obtain only 1 Mb/s data
transfer rate ).
with 3.6.x kernels (including the latest 3.6.4 kernel)
------------------------------------------------------
- a similar problem occurs, but performance falls less drastically :
we obtain a 25 Mb/s transfer rate.
with 3.2.x kernels
------------------
- no problem occurs ; we have good data transfers , even in 64 Gb
High Memory mode.
We measure data transfers by typing the following command:
dd if=/dev/zero of=./foobar bs=1000 count=10000
The distribution we are using for the tests is a just-installed i386
Debian 6.0 "squeeze".
When using some other drivers (such as the 'MPT fusion' driver) on some
other models
of servers, we have no problems .
Could you help us determining what goes wrong ... ?
Of course, we can send you any information (such as debug logs, kernel
.config files ...) you may need to do so ...
Have a nice day,
best regards
Jérémie GUILLAUME
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
