Start a Conversation

Unsolved

This post is more than 5 years old

1392

February 20th, 2008 02:00

Partition Alignment (once again)

Hi !

When setting up Partitions on a Lun under Linux (SLES9) and using the "Best Practices",
I have the following effects:

dd if=/dev/emcpowera of=/dev/null bs=64k gives about 300mb/s
dd if=/dev/encpowera1.... gives only 40mb/s, when not aligning.

The Partition uses the whole disk.
When now using the recommended way with fdisk nothing changes at all.
BUT when I leave some space by leaving some space at the end of the Lun
(creating partion not to the end of disk) and the use the recommended alignment the test
gives me again the whole 300mb/s.
The same is true for mke2fs. The filesystem creation is much faster when leaving space AND aligning.

I haven't found anything about leaving space is necessary so I think this might be unique to the clariion.

Any Ideas why this is happending ?

2 Intern

 • 

2.8K Posts

February 20th, 2008 03:00

I'm neither a Linux nor a Clariion expert .. But from the little i know about alignment, I can only say that the two DD you used tend to confirm that the right alignment (bs=64k) gives a leap in performances while an unaligned access (without the default bs) gives poor performances. When you use dd on a

Having that said, I need more informations about your environment.

Could you please specify the size of the lun. A good starting point may be the partition table as per the fdisk command for your CX lun. Please post both "aligned-only" and "aligned-and-padded" partition table.

I think you are using ext3 when formatting your lun. If possible please post also the mkfs command line you used.

16 Posts

February 20th, 2008 03:00

I should not be necessary to create the partition and change it with the expert mode of fdisk. It is much easiert just to toogle the units to sectors (u command) and create the partition directly startion at sector 128.

16 Posts

February 20th, 2008 04:00

A single disk gives about 80MB/s.
Read Performance with Raid5 (4+1) can be about 4 times higher.
And we do not write in this rather senseless test. writing to /dev/null just discards the data after reading.

2 Intern

 • 

5.7K Posts

February 20th, 2008 04:00

300MBps ? Waaw !

16 Posts

February 20th, 2008 04:00

To clarify: I meant Megabytes per second.
But 300 is not to much when just reading from disk without involving a filesystem.
I won't get this number with applications but I stumbled on this odd behavior I tried to understand.
The disks used a 10k FC 300GB.

2 Intern

 • 

5.7K Posts

February 20th, 2008 04:00

Not too much ? How many spindels are invoolved in this ? You are actually talking about a sustained throughput of 300 Megabytes per second ? (3Gbps) Waaw.

16 Posts

February 20th, 2008 04:00

What numbers did you expect ?

16 Posts

February 20th, 2008 04:00

Maybe you misinterpreted the dd.

The first gives Throughput on the whole lun, without accessing the Partition.
This works because the lun can't be misaligned.
When then creating a misaligned Partition and dd this, the performance is poor.

Creating a aligned partition gives the right throughput, but this does not work according to "best practices", where the default settings are used when creating the partition with fdisk.
Aligning only works when
- leaving space at the end of the lun
- using sector settings.

The 180gb Lun is on a raid5 with 5 Disks (128 element size).


Creating:
Disk /dev/emcpowere: 171.7 GB, 171798691840 bytes
255 heads, 63 sectors/track, 20886 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes

Device Boot Start End Blocks Id System

Command (m for help): n
Command action
e extended
p primary partition (1-4)
p
Partition number (1-4): 1
First cylinder (1-20886, default 1):
Using default value 1
Last cylinder or +size or +sizeM or +sizeK (1-20886, default 20886):
Using default value 20886

Command (m for help): x

Expert command (m for help): b
Partition number (1-4): 1
New beginning of data (63-335533589, default 63): 128

Expert command (m for help): p

Disk /dev/emcpowere: 255 heads, 63 sectors, 20886 cylinders

Nr AF Hd Sec Cyl Hd Sec Cyl Start Size ID
1 00 1 1 0 254 63 1023 128 335533462 83
2 00 0 0 0 0 0 0 0 0 00
3 00 0 0 0 0 0 0 0 0 00
4 00 0 0 0 0 0 0 0 0 00

This one is slow.

Now the working one:

Command (m for help): n
Command action
e extended
p primary partition (1-4)
p
Partition number (1-4): 1
First cylinder (1-20886, default 1): 1
Last cylinder or +size or +sizeM or +sizeK (1-20886, default 20886): 20800

Command (m for help): x

Expert command (m for help): b
Partition number (1-4): 1
New beginning of data (63-334151999, default 63): 128

Expert command (m for help): p

Disk /dev/emcpowere: 255 heads, 63 sectors, 20886 cylinders

Nr AF Hd Sec Cyl Hd Sec Cyl Start Size ID
1 00 1 1 0 254 63 1023 128 334151872 83
2 00 0 0 0 0 0 0 0 0 00
3 00 0 0 0 0 0 0 0 0 00
4 00 0 0 0 0 0 0 0 0 00

Expert command (m for help): w

The partions mostly look the same but dd is very different.

The easiest way to create a partition is:
Command (m for help): u
Changing display/entry units to sectors

Command (m for help): n
Command action
e extended
p primary partition (1-4)
p
Partition number (1-4): 1
First sector (63-335544319, default 63): 128
Last sector or +size or +sizeM or +sizeK (128-335544319, default 335544319):
Using default value 335544319

The dd is independent of filesystem.

2 Intern

 • 

5.7K Posts

February 20th, 2008 04:00

Depending on the number of spindels, I thought 50 or 100 would have been a nice value.

2 Intern

 • 

2.8K Posts

February 20th, 2008 05:00

I can see how ugly my post was ... Wooops .. Sorry for that .. It's unclear and poorly written .. Forgive me !!

2 Intern

 • 

5.7K Posts

February 20th, 2008 06:00

Senseless test... yeah.... It's not a real life scenario. you're only measuring the maximum throughput and you'll get a value you can only hope to see in the future when you're doing production on the Clariion.

aaaah, that makes sence.

2 Intern

 • 

5.7K Posts

February 20th, 2008 06:00

/slap

2 Intern

 • 

2.8K Posts

February 20th, 2008 06:00

Rob, Omega .. I think that this question may be better answered if moved in the Clariion area of the forums .. What do you think ??

If you agree, I'll ask our admins to move this thread .. and you'll hopefully have a lot more interest on this topic :D

2 Intern

 • 

5.7K Posts

February 20th, 2008 07:00

Well, it began as a host rrelated thingy, but grew towards a Clariion related topic. I don't know... If we leave it at this, we could leave it here.

Omega, what do you think ?

16 Posts

February 20th, 2008 22:00

Yes, please move it.
No Events found!

Top