Jump to content


Photo

Hardware RAID and AF 4k with XP


  • You cannot start a new topic
  • Please log in to reply
8 replies to this topic

#1 Bibeu

Bibeu

    Member

  • Member
  • 11 posts

Posted 17 March 2013 - 12:26 PM

Hi
I post here because I can't get help in my language.
I have a Promise EX8350
I bought 5 disks hgst 7K1000.D 1TB to replace my system (C:\) 500GB RAID10 array (1 dyn-spare disk in the end for my intended 2TB RAID10 LD).
Please could one advise about RAID sector size and related issue(s) when creating the LD, knowing my best wish is to clone my current system I already cloned to one of the 5 new disks I attached to the mobo (for security purpose and to make room on the controller board that hosts another small array raid1 for data).
The controller firmware allows 512b/1kB/2kB/4kB sectors. Is this some low-level format or can this break the native 4k sectors?
I already tested a cloning to a 512b sectors array which works very slowly and is unaligned. I launched a full sync to the array but this is very looooong 1%/h which is ~6MB/s and will end tonight in 7 hours (93% elapsed).
When I pause the sync and I run HDtune to the array it reports horrible curve up to 15% with throughput picks between 10~50MB/s. Then above 15% to the end a nearly flat line at ~125MB/s and some peaks down to 40-110MB/s with 64k test blocks, and a lot of peaks down to 100MB/s when testing with 8MB blocks.
Neither Google/Wikipedia searches answered my questions.
Thank you for your lights

#2 continuum

continuum

    Mod

  • Mod
  • 3,518 posts

Posted 18 March 2013 - 04:06 PM

The controller firmware allows 512b/1kB/2kB/4kB sectors. Is this some low-level format or can this break the native 4k sectors?

Most harddisks should either be 512-byte sectors or 4K (AFF) sectors. I am not sure where 1K/2K sectors would come in unfortunately-- does your controller manual clarify this at all?

Honestly 1%/hour is pretty slow, but the EX8350 is an ancient card by today's standards so that might be realistic...

#3 Bibeu

Bibeu

    Member

  • Member
  • 11 posts

Posted 19 March 2013 - 03:15 AM

Most harddisks should either be 512-byte sectors or 4K (AFF) sectors. I am not sure where 1K/2K sectors would come in unfortunately-- does your controller manual clarify this at all?

Honestly 1%/hour is pretty slow, but the EX8350 is an ancient card by today's standards so that might be realistic...

Thank you for replying continuum
The raid is now fully initialized and booted from. I resized the partition *before* realigning it, good idea because PQ PartitionMagic will exit on error when running from an align disk ("Can't detect volume letter").
Although, the performance curves are the same they were before init/resize/align.
As I can't get help anywhere I will now delete it and recreate with a 4k sector size, just hoping this is not just a stupid idea and so a time waste.
And yes the controller firmware allows to create 512B/1K/2K/4K sectors logical disk, but not to change this after creation.
Do you think 4k is the better choice? My idea is that I won't need to realign after cloning and this will spare calculations in the HBA because NTFS will use the default 4k clusters. More with 1k... 63*1024/4096=15.75 and 2k 63*2048/4096=31.5 won't result in an aligned partition.
Pictures below are with 512B sectors
[EDIT]
Seems I have a defective drive: I broke the RAID and tested them one by one and found one that showed the same erratic curve up to 25% (twice the RAID10 percent because drive size is half the RAID). Promise RAID-10 is a mirror of RAID-0

Attached Files


Edited by Bibeu, 20 March 2013 - 09:49 AM.

#4 continuum

continuum

    Mod

  • Mod
  • 3,518 posts

Posted 20 March 2013 - 12:31 PM

Hmm, it does look like you have something slow or out of whack on that drive.

And yes the controller firmware allows to create 512B/1K/2K/4K sectors logical disk, but not to change this after creation.
Do you think 4k is the better choice?

Modern harddisks are AFF (4K sectors), if you have AFF disks, 4K sectors probably make the most sense. However I'm still suspicious that that value in the RAID controller isn't quite what I'm thinking it's for-- usually stripe size is referred to as stripe size and not "sector size" unless we are losing something in your translation?

#5 Bibeu

Bibeu

    Member

  • Member
  • 11 posts

Posted 20 March 2013 - 04:01 PM

Hmm, it does look like you have something slow or out of whack on that drive.

Modern harddisks are AFF (4K sectors), if you have AFF disks, 4K sectors probably make the most sense. However I'm still suspicious that that value in the RAID controller isn't quite what I'm thinking it's for-- usually stripe size is referred to as stripe size and not "sector size" unless we are losing something in your translation?

No confusion neither lost in translation ;)

Attached Files


#6 continuum

continuum

    Mod

  • Mod
  • 3,518 posts

Posted 21 March 2013 - 01:41 PM

Then in that case I would make sure the sector size is set to match whatever your harddisks natively use (either 512 bytes or 4KB), and ignore the 1KB and 2KB settings as those are not sizes you are likely to ever encounter on today's harddisks. You may experience severe performance penalties if you try them. (or you may not, but why take the risk?)

#7 Bibeu

Bibeu

    Member

  • Member
  • 11 posts

Posted 22 March 2013 - 05:05 AM

They are 512e AF drives. I tried 4k and XP fsutil fsinfo ntfsutil reports 3 4096 fields (XP won't show an explicit 4096 physical sector size). The SuperBuild utility (pre boot BIOS overlay software) reports 4k sectors although the WebPAM management software keeps stuck showing 512B sectors.
When 4k initialized I get issues with HD sofwares like HDtune or PartitionInfo reporting wrong size eg 250Go for a 2TB LD.
OK I'll have the defective drive replaced and keep at 512B and clone then align.

Two last questions please, to help me choose the good RAID-0+1 stripe size:
When writing a file smaller than stripe size, is the whole stripe rewriten to the disk?
Can more than one file be stored in a stripe as long as their cumulated size is lower than stripe size?
Thank you for being there continnuum

Edited by Bibeu, 22 March 2013 - 06:06 AM.

#8 sub.mesa

sub.mesa

    Member

  • Member
  • 25 posts

Posted 03 April 2013 - 05:23 PM

When 4k initialized I get issues with HD sofwares like HDtune or PartitionInfo reporting wrong size eg 250Go for a 2TB LD.

That is a bug in HDtune; it blindly assumes disks have 512 byte sectors. It is shitty software. It is not the only thing that is bad about this piece of software. It confused many people believing something is wrong (i.e. with performance figures).

Stripesize is something very much misunderstood as well. Writing a file smaller than the stripesize will not cause the entire stripe block to be written. Stripesize is only a mechanism to determine what data ends up on which drive. In most cases, the stripesize can only be too small, not too large. Some users use stripesizes of up to 16 megabytes. Popular belief has it that this causes very weak performacne on 4K random I/O; but the opposite is true: larger stripesizes favor smaller I/O or random I/O - i.e. small files, while smaller stripesizes are better for sustained sequential I/O like large files.

#9 Bibeu

Bibeu

    Member

  • Member
  • 11 posts

Posted 05 April 2013 - 08:59 AM

Stripesize is something very much misunderstood as well. Writing a file smaller than the stripesize will not cause the entire stripe block to be written. Stripesize is only a mechanism to determine what data ends up on which drive. In most cases, the stripesize can only be too small, not too large. Some users use stripesizes of up to 16 megabytes. Popular belief has it that this causes very weak performacne on 4K random I/O; but the opposite is true: larger stripesizes favor smaller I/O or random I/O - i.e. small files, while smaller stripesizes are better for sustained sequential I/O like large files.

Thank you sub.mesa
So assuming my controller is allowing only 32-64-128k stripes and a desktop usage with sometimes video/music handling and everyday web-browsing (think to the cache) and mail client, I should decide for 128k?
Although when a bad sector is found on a drive pertaining to a RAID0+1, won't the whole stripe be flagged as some BSL entry, so a big stripe will cause more lost space?
And just to be sure, won't a 128k stripe will end in 124k space lost when writing a small 4k file (eg a windows shortcut or gnome desktop launcher or small text/image file)?



0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users