Comments on: Why we need 4k drives

By: Brian’s Brain » Blog Archive » To Infinity And Beyond: Seagate Turns 3 TByte HDDs On

Wed, 07 Jul 2010 17:29:25 +0000

[...] mitigating the degrading effects of raw error rates. Unfortunately, the 4 Kbyte sector size is at minimum performance- and capacity-inefficient, and worst-case completely incompatible, with legacy operating systems when the drive is directly accessed by them. By putting an [...]

By: Chris Ferebee, Frankfurt» Blogarchiv » 1 TB im MacBook Pro

Chris Ferebee, Frankfurt» Blogarchiv » 1 TB im MacBook Pro — Fri, 02 Apr 2010 12:58:55 +0000

[...] Ich verwende die Western Digital WD10TPVT, nachdem die ursprünglich angekündigte WD10TEVT nie in Stückzahlen ausgeliefert wurde. Die WD10TPVT unterscheidet sich durch eine leicht geringere Umdrehungsgeschwindigkeit (5200 UpM statt 5400) und der neuen 4K-Sektorgröße, die für moderne Betriebssysteme nur Vorteile bringt, z. B. eine bessere Fehlerkorrektur. [...]

By: Block alignment is critical | CTI Strategy Blog

Block alignment is critical | CTI Strategy Blog — Fri, 26 Mar 2010 23:18:43 +0000

[...] that is often overlooked in storage. I read a blog entry by Robin Harris a couple months back about the importance of block alignment with the new 4KB drives. I was curious to test the theory on one of the new 4KB drives, but I did not have one on hand. [...]

By: Alvin

Alvin — Thu, 11 Mar 2010 08:04:42 +0000

Actually, using 4k-sector format drives is strongly motivated by two major persectives which the conventional 512Byte (half-k sector format) drive is far lagged behand:

1. Disk format efficiency improvement by about 10% capacity increase – each sector starts with disk preamble data that has no user information but presents least information necessary for disk manufactorers. using larger sector format, i.e. 4k sector format, has saved 7 such gap space for storing user data per 4096-Byte sector that the conventional 512Byte sector drive cannot save, which largely yields extra recording spaceds for user, given the same density is applied. In WD drives, users should not notice any difference.
2. Using 4k sector format drive not only increases the conventional RS-ECC correction capability, but is potentially replacing the RS-ECC with the new revolutional LDPC code technology, which will provide significant disk drive URE performance. For instance, given URE of every 1 of 10^14 bit data read in conventional RS-ECC, by using LDPC and 4k sector format, the drive can easily go down to every 1 of 10^22 URE. This will translate into a huge benefit to users, Os and HDD industries with new technologies in head and media as well. In the end, parameers such as AFR, SMART and MTBF will be largely improved. RAID-x configuration may be rethought with these new gerneration high-performance drives.

By: IBM legacy BIOS

IBM legacy BIOS — Sun, 27 Dec 2009 05:42:23 +0000

The link provided by Karl makes a valid point about DOS/BIOS based partition tables for x86. An issue some SAN/VMWare environments suffer from as well.

This misalignment can cut peak IOPs in half or more.

Still waiting for EFI ( http://en.wikipedia.org/wiki/Extensible_Firmware_Interface ) motherboards with GUID partition table support (http://en.wikipedia.org/wiki/GUID_Partition_Table ) to take off

By: Karl Katzke

Karl Katzke — Thu, 24 Dec 2009 04:29:37 +0000

The 4k thing is also sort of screwing over Drobo users too... funny that I saw these two posts on the same day!

By: Drobo, XP Users: Beware of 4K “Advanced Format” Drives! – Stephen Foskett, Pack Rat

Wed, 23 Dec 2009 21:55:29 +0000

[...] StorageMojo: Why we need 4k drives [...]

By: gonzio

gonzio — Wed, 23 Dec 2009 16:31:47 +0000

Clariion uses 520 bytes sectors, but 8 of those bytes are used internally by the storage to store a time stamp and a checksum to ensure data integrity. The host only sees the 512 bytes sectors, so it is the same as with internal disks.

By: Eric

Eric — Wed, 23 Dec 2009 15:32:27 +0000

The presumption of this is that existing data will be reformatted to 4k sector file systems, and that 512 emulation on 4k will offer as good or better performance. Both are huge assumptions not likely to be true at the 90% level for years.

By: Rob

Rob — Wed, 23 Dec 2009 02:51:46 +0000

Yes… Clariion uses 520 byte sectors. Secondly, UBE of Enterprise FC and SATA is 1 in 10 to the 15th, not 1 in 10 to the 14th. That’s a *real* important distinction… If the 400 bytes of ECC on 4096 byte sector greatly increases reliability, I’m all for it. The extra 8 bytes on 512 bytes, hasn’t impressed me

By: Martin K. Petersen

Martin K. Petersen — Tue, 22 Dec 2009 18:46:50 +0000

Linux has supported devices with native 4KB sectors for a long time. As far as drives with 512-byte logical and 4KB physical sectors is concerned the support was included in the 2.6.31 kernel. Our partitioning and LVM tools have been updated to compensate for the alignment reported by drive (ATA and SCSI). I think Fedora is the only distribution that ships the relevant bits at this point.

It is hard for a general purpose operating system to deal with buffer sizes that are not multiples of 512 in an efficient manner. So 520/528-byte sector drives have mostly been used inside RAID arrays. The extra bytes are used to store information proprietary to the array firmware.

Linux does support 520/4106-byte sector drives, but only when they are formatted using T10 Protection Information (DIF) and hanging off a DIF or DIX-capable HBA.

By: Dali

Dali — Tue, 22 Dec 2009 16:35:27 +0000

Great in depth article from Anandtech covering the need for 4K sectors, the issues and the workarounds.

http://www.anandtech.com/storage/showdoc.aspx?i=3691

By: holger

holger — Tue, 22 Dec 2009 09:07:33 +0000

I think several vendors of raid boxes use a 520 byte formatting *internally* to store extra metadata there.
Even good old VMS presented 512 bytes to the user but used the additional bytes for storing the “forced error flag”.

By: nate

nate — Mon, 21 Dec 2009 18:51:34 +0000

It seems EMC uses 520 bytes, not sure if anyone else does – http://chucksblog.emc.com/chucks_blog/2008/08/your-storage-mi.html

Also I think the file system block size is related? In that if your drive has 512-byte sectors and your file system is using 4096-byte blocks then you’ll probably be more efficient with a 4k physical sector? I don’t recall the last time linux was able to use a 512-byte file system block size(man page says supported values are 1-4k), I think the default with most larger file system sizes is 4k today already(on linux at least).

On my big storage array the average I/O size is about 37kB( to disk, over 100kB to cache).