LWN: Comments on "LogFS" https://lwn.net/Articles/234441/ This is a special feed containing comments posted to the individual LWN article titled "LogFS". en-us Mon, 06 Oct 2025 18:54:22 +0000 Mon, 06 Oct 2025 18:54:22 +0000 https://www.rssboard.org/rss-specification lwn@lwn.net LogFS https://lwn.net/Articles/240970/ https://lwn.net/Articles/240970/ crucialp This is a very interesting idea, and great to see how it could be included in mainline.<br> Mon, 09 Jul 2007 02:20:22 +0000 LogFS https://lwn.net/Articles/235593/ https://lwn.net/Articles/235593/ Duncan So we have a "wandering journal", much like reiser4 is supposed to have. <br> What would be the possibility of adding a plugin to reiser4 to make it <br> behave with flash requirements in mind as opposed to rotating disc <br> requirements?<br> <p> Of course, reiser4 isn't in mainline yet, but supposedly, it's getting <br> closer, and with Namesys employees continuing to work toward that, and <br> with Hans himself (fs genius but political sand in the gears he often <br> seemed to be) out of the equation now, merging for 2.6.23 or .24 is <br> beginning to look both politically and technically possible. (See this <br> article from the April 26 LWN kernel page, "Filesystems: chunkfs and <br> reiser4", <a rel="nofollow" href="http://lwn.net/Articles/231585/">http://lwn.net/Articles/231585/</a> .)<br> <p> Duncan<br> Thu, 24 May 2007 11:03:00 +0000 FTL and obsolete sectors https://lwn.net/Articles/235020/ https://lwn.net/Articles/235020/ seanyoung The logfs paper mentions obsoleting data in an ftl. I wrote a patch for that some time back but it didn't work all that well.<br> <p> <a href="http://lwn.net/Articles/162776/">http://lwn.net/Articles/162776/</a><br> <p> I introduced a new barrier request "forget" (obsolete would have been a better name). The ftl block device can used this information; however the barrier request caused more writes than without, as write requests before and after the barrier cannot be merged.<br> <p> For example, the fat table itself is repeatedly written. If:<br> <p> - write fat<br> - forget barrier request<br> - write fat again<br> <p> Without the forget barrier the first write can be merged with second. So the only way to do this properly is by adding functionality to the block layer which can merge forget/obsolete requests. So the patch as describes actually increases I/O.<br> <p> Another problem was that the in-kernel FTL layers are not used much nowadays, so the benefit would be limited. OTOH, the CompactFlash ATA command set does have an "erase" command which could possibly do exactly this -- I never verified this. However if it does compact flash memory would be faster and last longer.<br> Fri, 18 May 2007 20:32:16 +0000