Support discarding of unused sectors - Revision history

J3gum: Update status of the work, and add a link to FITRIM/discard, minor formatting, spell check

2011-10-11T19:42:26Z

Update status of the work, and add a link to FITRIM/discard, minor formatting, spell check

← Older revision		Revision as of 19:42, 11 October 2011
Line 3:		Line 3:
	For magnetic disks, DRAM and many other physical media used as block devices, it does not matter for the device to know which sectors are actually unused by the filesystem and therefore contain irrelevant content.		For magnetic disks, DRAM and many other physical media used as block devices, it does not matter for the device to know which sectors are actually unused by the filesystem and therefore contain irrelevant content.

	This is quite different for Solid State Disks (SSDs) built from Flash Memory: To avoid performance degradation from the dynamic wear-~~levelling~~ algorithms, there has to be a clever management of "unused" blocks in the SSD controller.		This is quite different for Solid State Disks (SSDs) built from Flash Memory: To avoid performance degradation from the dynamic wear-leveling algorithms, there has to be a clever management of "unused" blocks in the SSD controller.

	To let the SSD controller know which sectors contain irrelevant data, the ATA TRIM command has been standardized, and now the first SSDs are available that implement that command (e.g. the SSDs using the Indilinx Barefoot controller, from firmware revision 1370 onwards).		To let the SSD controller know which sectors contain irrelevant data, the ATA TRIM command has been standardized, and now the first SSDs are available that implement that command (e.g. the SSDs using the Indilinx Barefoot controller, from firmware revision 1370 onwards).
Line 9:		Line 9:
	== What needs to be done ==		== What needs to be done ==

	~~To support SSDs and other devices that benefit from knowing unused sectors, the blkdev_issue_discard() function~~ has ~~recently~~ been ~~integrated into the~~ Linux ~~kernel (2~~.6.~~30rcX).~~		This work has been implemented in Linux since v2.6.38, Jan 18 2011.

	~~The filesystem needs to call that function when it knows sectors become unused.~~

	~~Linux filesystems known to already support blkdev_issue_discard() (when available) are: ext4, GFS2~~, ~~swap~~.

	== Related Standards ==		== Related Standards ==

	[http://www.t13.org/Documents/UploadedDocuments/docs2008/e07154r6-Data_Set_Management_Proposal_for_ATA-ACS2.doc Data Set Management Commands Proposal for ATA8-ACS2]		* [http://www.t13.org/Documents/UploadedDocuments/docs2008/e07154r6-Data_Set_Management_Proposal_for_ATA-ACS2.doc Data Set Management Commands Proposal for ATA8-ACS2]
			* [http://www.t13.org/Documents/UploadedDocuments/docs2008/e08137r4-DRAT_-_Deterministic_Read_After_Trim.doc Deterministic TRIM Proposal for ATA8-ACS2]
	[http://www.t13.org/Documents/UploadedDocuments/docs2008/e08137r4-DRAT_-_Deterministic_Read_After_Trim.doc Deterministic TRIM Proposal for ATA8-ACS2]

	== Other Documents ==		== Other Documents ==

			* [[FITRIM/discard]]
	* An interesting [http://www.anandtech.com/printarticle.aspx?i=3531 Article at Anandtech] that (while being a little verbose and Windows-oriented) explains many aspects of SSD performance, including the prospective benefits of the ATA TRIM command (at page 10).		* An interesting [http://www.anandtech.com/printarticle.aspx?i=3531 Article at Anandtech] that (while being a little verbose and Windows-oriented) explains many aspects of SSD performance, including the prospective benefits of the ATA TRIM command (at page 10).

	* Patch [http://www.mail-archive.com/cluster-devel@redhat.com/msg03401.html adding the blkdev_issue_discard()] feature to GFS2		* Patch [http://www.mail-archive.com/cluster-devel@redhat.com/msg03401.html adding the blkdev_issue_discard()] feature to GFS2

	* [http://www.usenix.org/event/lsf08/tech/shin_SSD.pdf Presentation] on a comparison between different filesystems performance on SSDs (including XFS)		* [http://www.usenix.org/event/lsf08/tech/shin_SSD.pdf Presentation] on a comparison between different filesystems performance on SSDs (including XFS)

Pniemayer: /* Other Documents */

2009-04-24T12:51:04Z

Pniemayer: /* What needs to be done */

2009-04-17T16:05:20Z

What needs to be done

← Older revision		Revision as of 16:05, 17 April 2009
Line 12:		Line 12:

	The filesystem needs to call that function when it knows sectors become unused.		The filesystem needs to call that function when it knows sectors become unused.

			Linux filesystems known to already support blkdev_issue_discard() (when available) are: ext4, GFS2, swap.

	== Related Standards ==		== Related Standards ==

Pniemayer: + links to related documents

2009-04-15T12:27:59Z

+ links to related documents

← Older revision		Revision as of 12:27, 15 April 2009
Line 18:		Line 18:

	[http://www.t13.org/Documents/UploadedDocuments/docs2008/e08137r4-DRAT_-_Deterministic_Read_After_Trim.doc Deterministic TRIM Proposal for ATA8-ACS2]		[http://www.t13.org/Documents/UploadedDocuments/docs2008/e08137r4-DRAT_-_Deterministic_Read_After_Trim.doc Deterministic TRIM Proposal for ATA8-ACS2]

			== Other Documents ==

			* An interesting [http://www.anandtech.com/printarticle.aspx?i=3531 Article at Anandtech] that (while being a little verbose and Windows-oriented) explains many aspects of SSD performance, including the prospective benefits of the ATA TRIM command (at page 10).

			* Patch [http://www.mail-archive.com/cluster-devel@redhat.com/msg03401.html adding the blkdev_issue_discard()] feature to GFS2

Pniemayer: initial proposal

2009-04-15T11:55:57Z

initial proposal

New page

== Rationale ==

For magnetic disks, DRAM and many other physical media used as block devices, it does not matter for the device to know which sectors are actually unused by the filesystem and therefore contain irrelevant content.

This is quite different for Solid State Disks (SSDs) built from Flash Memory: To avoid performance degradation from the dynamic wear-levelling algorithms, there has to be a clever management of "unused" blocks in the SSD controller.

To let the SSD controller know which sectors contain irrelevant data, the ATA TRIM command has been standardized, and now the first SSDs are available that implement that command (e.g. the SSDs using the Indilinx Barefoot controller, from firmware revision 1370 onwards).

== What needs to be done ==

To support SSDs and other devices that benefit from knowing unused sectors, the blkdev_issue_discard() function has recently been integrated into the Linux kernel (2.6.30rcX).

The filesystem needs to call that function when it knows sectors become unused.

== Related Standards ==

[http://www.t13.org/Documents/UploadedDocuments/docs2008/e07154r6-Data_Set_Management_Proposal_for_ATA-ACS2.doc Data Set Management Commands Proposal for ATA8-ACS2]

[http://www.t13.org/Documents/UploadedDocuments/docs2008/e08137r4-DRAT_-_Deterministic_Read_After_Trim.doc Deterministic TRIM Proposal for ATA8-ACS2]

← Older revision		Revision as of 12:51, 24 April 2009
Line 26:		Line 26:

	* Patch [http://www.mail-archive.com/cluster-devel@redhat.com/msg03401.html adding the blkdev_issue_discard()] feature to GFS2		* Patch [http://www.mail-archive.com/cluster-devel@redhat.com/msg03401.html adding the blkdev_issue_discard()] feature to GFS2

			* [http://www.usenix.org/event/lsf08/tech/shin_SSD.pdf Presentation] on a comparison between different filesystems performance on SSDs (including XFS)