Talk:Debian Jessie on XO: Difference between revisions

From OLPC
Jump to navigation Jump to search
No edit summary
m (→‎Ext4 on an SD Card: Further relies.)
Line 18: Line 18:
:Disabling the journal will damage your filesystem when the XO-1 runs out of battery power and is not plugged in. Batteries made in 2007 or 2008 should be very ready to fail. If you need I/O performance, you should not be using an XO-1. The XO-1 design is from 2006. A Raspberry Pi 3 outperforms an XO-1 by an order of magnitude or two. Our kernel is after that blog post by three years, (2013 vs 2010), so I don't think the post is of use except to suggest things to test if a system is I/O bandwidth constrained. You must have a performance measurement method, and use it before and after the change, to see what the effect is. So what is your measurement method? The best way to reduce I/O delay on the XO-1 is to remove I/O demand; size the application demand to fit the available memory (256 MB), so that enough of the filesystem metadata is in cache. Exceed the available memory and the system will begin to discard and then re-read metadata. To reduce the application demand, avoid running programs that are not needed, and do not run a web browser. Hardly practical with today's focus on web sites, but I've no idea what your overall use for your XO-1 is. --[[User:Quozl|Quozl]] 21:46, 16 January 2017 (UTC)
:Disabling the journal will damage your filesystem when the XO-1 runs out of battery power and is not plugged in. Batteries made in 2007 or 2008 should be very ready to fail. If you need I/O performance, you should not be using an XO-1. The XO-1 design is from 2006. A Raspberry Pi 3 outperforms an XO-1 by an order of magnitude or two. Our kernel is after that blog post by three years, (2013 vs 2010), so I don't think the post is of use except to suggest things to test if a system is I/O bandwidth constrained. You must have a performance measurement method, and use it before and after the change, to see what the effect is. So what is your measurement method? The best way to reduce I/O delay on the XO-1 is to remove I/O demand; size the application demand to fit the available memory (256 MB), so that enough of the filesystem metadata is in cache. Exceed the available memory and the system will begin to discard and then re-read metadata. To reduce the application demand, avoid running programs that are not needed, and do not run a web browser. Hardly practical with today's focus on web sites, but I've no idea what your overall use for your XO-1 is. --[[User:Quozl|Quozl]] 21:46, 16 January 2017 (UTC)


<br>
:"Disabling the journal will damage your filesystem ..." <br>
:"Disabling the journal will damage your filesystem ..." <br>
::Whereas journalling is claimed to shorten the life of the SD card. Computers were used successfully with ext2 and fsck. Is journalling necessary?<br>
::Whereas journalling is claimed to shorten the life of the SD card. Computers were used successfully with ext2 and fsck. Is journalling necessary?<br>
:::Journalling will not be a primary cause of shortened life; it will be the read-disturb-writes. Your concern about the life of the card is misplaced; the card cost is trivial, and there's no advantage to using the more costly higher speed cards because the controller does not support the higher speed modes. fsck is bad because it causes even more read-disturb-writes, so it should be avoided. Journalling is only used when you write files; which for ''a little text editing and web access'' means not much. But there are operating system logs too. On OLPC OS we removed all unnecessary file writes, but when you switch to Debian you have lost all those optimisations. At this stage of the product life, your battery should fail before your SD card does. --[[User:Quozl|Quozl]] 02:02, 18 January 2017 (UTC)
:::Journalling will not be a primary cause of shortened life; it will be the read-disturb-writes. Your concern about the life of the card is misplaced; the card cost is trivial, and there's no advantage to using the more costly higher speed cards because the controller does not support the higher speed modes. fsck is bad because it causes even more read-disturb-writes, so it should be avoided. Journalling is only used when you write files; which for ''a little text editing and web access'' means not much. But there are operating system logs too. On OLPC OS we removed all unnecessary file writes, but when you switch to Debian you have lost all those optimisations. At this stage of the product life, your battery should fail before your SD card does. --[[User:Quozl|Quozl]] 02:02, 18 January 2017 (UTC)

::::Yes, the purchase cost of an SD card is negligible. The significant cost is restoration of data after a card fails. I run fsck only prior to mounting. Typically the volume is mounted only when the software needing it is started. So fsck is much less frequent than file write. In my experience, an occasional crash of an application is unavoidable. Mounting of a broken file system must be avoided. To my knowledge there are three possibilities. (1) After a crash, rely on journaling to take care of the file system. (2) After a crash when a file is open, apply fsck and allow it to attempt repairs. (3) After any crash of software with an open file, restore the file system from a backup. As I understand, you prefer (1) whereas the advice in the introduction in https://en.wikipedia.org/wiki/Ext2 necessitates (2) or (3). Regards, ... [[User:Peasthope|Peasthope]] 17:57, 21 January 2017 (UTC)

:"... when the XO-1 runs out of battery power and is not plugged in." <br>
:"... when the XO-1 runs out of battery power and is not plugged in." <br>
::With a properly working power system, the low battery warning allows clean shutdown. Correct?<br>
::With a properly working power system, the low battery warning allows clean shutdown. Correct?<br>
:::I'm speaking from the point of view of scalability. There will be laptops that do not cleanly shutdown, because it is impossible to know the state of charge of the battery. Capacity value displayed is an estimate. As your battery ages, this warning will tend not to occur at the correct time. OLPC OS has a properly working system for the XO-1.5, but Debian does not; we have powerd version 113 in OLPC OS, but Debian has a very old version 23 which pre-dates XO_1.5. --[[User:Quozl|Quozl]] 02:02, 18 January 2017 (UTC)
:::I'm speaking from the point of view of scalability. There will be laptops that do not cleanly shutdown, because it is impossible to know the state of charge of the battery. Capacity value displayed is an estimate. As your battery ages, this warning will tend not to occur at the correct time. OLPC OS has a properly working system for the XO-1.5, but Debian does not; we have powerd version 113 in OLPC OS, but Debian has a very old version 23 which pre-dates XO_1.5. --[[User:Quozl|Quozl]] 02:02, 18 January 2017 (UTC)

::::So considering system optimizations and powerd, forget Debian and stick to OLPC. Regards, ... [[User:Peasthope|Peasthope]] 17:57, 21 January 2017 (UTC)

:"XO-1" <br>
:"XO-1" <br>
::XO-1.5 here.
::XO-1.5 here.

Revision as of 17:57, 21 January 2017

last?

In the History page,

2015-11-26T07:01:56 Quozl (Talk | contribs) (6,793 bytes) (using last instead of ext, ...

last is a synonym for ZFS? Another file system? A typo? Thanks, ... Peasthope 05:31, 23 December 2016 (UTC)

Nothing to do with filesystem type; firmware detects that automatically. No, not a typo. Did you look at the difference between revisions to understand the edit? ext and last are device name alias in the firmware, see bootparam.fth for how last is defined. --Quozl 18:33, 23 December 2016 (UTC)
Thanks. I was tired and thinking file systems rather than Forth.

Ext4 on an SD Card

Instructions include "An SD card ... formatted to ext4 ...". According to most discussions, best current practice with flash storage is ext4 without journaling. Options for file system creation and mounting here.
http://fenidik.blogspot.ca/2010/03/ext4-disable-journal.html
Any additional tips? Thanks ....Peasthope 18:04, 16 January 2017 (UTC)

Disabling the journal will damage your filesystem when the XO-1 runs out of battery power and is not plugged in. Batteries made in 2007 or 2008 should be very ready to fail. If you need I/O performance, you should not be using an XO-1. The XO-1 design is from 2006. A Raspberry Pi 3 outperforms an XO-1 by an order of magnitude or two. Our kernel is after that blog post by three years, (2013 vs 2010), so I don't think the post is of use except to suggest things to test if a system is I/O bandwidth constrained. You must have a performance measurement method, and use it before and after the change, to see what the effect is. So what is your measurement method? The best way to reduce I/O delay on the XO-1 is to remove I/O demand; size the application demand to fit the available memory (256 MB), so that enough of the filesystem metadata is in cache. Exceed the available memory and the system will begin to discard and then re-read metadata. To reduce the application demand, avoid running programs that are not needed, and do not run a web browser. Hardly practical with today's focus on web sites, but I've no idea what your overall use for your XO-1 is. --Quozl 21:46, 16 January 2017 (UTC)
"Disabling the journal will damage your filesystem ..."
Whereas journalling is claimed to shorten the life of the SD card. Computers were used successfully with ext2 and fsck. Is journalling necessary?
Journalling will not be a primary cause of shortened life; it will be the read-disturb-writes. Your concern about the life of the card is misplaced; the card cost is trivial, and there's no advantage to using the more costly higher speed cards because the controller does not support the higher speed modes. fsck is bad because it causes even more read-disturb-writes, so it should be avoided. Journalling is only used when you write files; which for a little text editing and web access means not much. But there are operating system logs too. On OLPC OS we removed all unnecessary file writes, but when you switch to Debian you have lost all those optimisations. At this stage of the product life, your battery should fail before your SD card does. --Quozl 02:02, 18 January 2017 (UTC)
Yes, the purchase cost of an SD card is negligible. The significant cost is restoration of data after a card fails. I run fsck only prior to mounting. Typically the volume is mounted only when the software needing it is started. So fsck is much less frequent than file write. In my experience, an occasional crash of an application is unavoidable. Mounting of a broken file system must be avoided. To my knowledge there are three possibilities. (1) After a crash, rely on journaling to take care of the file system. (2) After a crash when a file is open, apply fsck and allow it to attempt repairs. (3) After any crash of software with an open file, restore the file system from a backup. As I understand, you prefer (1) whereas the advice in the introduction in https://en.wikipedia.org/wiki/Ext2 necessitates (2) or (3). Regards, ... Peasthope 17:57, 21 January 2017 (UTC)
"... when the XO-1 runs out of battery power and is not plugged in."
With a properly working power system, the low battery warning allows clean shutdown. Correct?
I'm speaking from the point of view of scalability. There will be laptops that do not cleanly shutdown, because it is impossible to know the state of charge of the battery. Capacity value displayed is an estimate. As your battery ages, this warning will tend not to occur at the correct time. OLPC OS has a properly working system for the XO-1.5, but Debian does not; we have powerd version 113 in OLPC OS, but Debian has a very old version 23 which pre-dates XO_1.5. --Quozl 02:02, 18 January 2017 (UTC)
So considering system optimizations and powerd, forget Debian and stick to OLPC. Regards, ... Peasthope 17:57, 21 January 2017 (UTC)
"XO-1"
XO-1.5 here.
Debian interests me for these reasons.
  • I use Debian on other machines. Debian on the XO-1.5 is one less system to contend with.
  • For package management, I prefer aptitude over yum.
  • In my experience, configuration of TLS is automated better in Debian than in stock OLPC. I might have missed a helpful package.
"So what is your measurement method?"
Performance is a secondary concern. If Debian is noticeably slower than OLPC on the XO-1.5, I'll have to consider the trade-offs.
"A Raspberry Pi 3 outperforms an XO-1 by an order of magnitude or two."
Worth noting. A http:://wiki.laptop.org/Performance_comparisons page cou‎ld be worthwhile. I'm not aware of any other machine with housing, reliability, ruggedness, screen legibility and water resistance competing with the XO.
No thanks, I'm not offering to maintain such a page; no audience, and no benefit. --Quozl 02:02, 18 January 2017 (UTC).
"I've no idea what your overall use for your XO-1 is."
Desktop systems work well at home or office. The XO is well suited for travel. A little text editing and Web access. Nothing sophisticated. The XO is also helpful for reading documentation when a desktop machine is out of commission.
Regards, ... Peasthope 18:43, 17 January 2017 (UTC)