| |||||||||||||
|
| |||||||||||||
First use of UBIFS?First use of UBIFS?Posted Oct 10, 2008 12:46 UTC (Fri) by dwmw2 (subscriber, #2063)In reply to: First use of UBIFS? by epa Parent article: The 2.6.27 kernel is out UBIFS is for use on real flash. SSD is not real flash. Well, it might be inside, but that's none of your business. It has a microcontroller between you and the flash, running a kind of pseudo-filesystem of its own, making it pretend to be a hard drive. So far, I've never seen an SSD where the code on that internal microcontroller is actually any good, and where it won't lose data if you subject it to stuff like power-fail testing. Who knows? Maybe one day they'll be suitable for real data. Maybe some of them already are? And that's ignoring the efficiency and the wear-levelling implications of running one journalling file system on top of another. The new support we have for telling the underlying block device that it can forget the contents of certain sectors will help — but really, we can do better if we just let the dog see the rabbit. Give us real flash, and get your steenking microcontrollers out of our way.
(Log in to post comments)
First use of UBIFS? Posted Oct 10, 2008 13:09 UTC (Fri) by dwmw2 (subscriber, #2063) [Link]
I should probably point out that I haven't been blessed with one of the shiny new drives that Linus raves about yet. Those are still like gold dust, it seems. So I haven't done any testing on those, and they're not included in the above observations.
First use of UBIFS? Posted Oct 10, 2008 13:14 UTC (Fri) by epa (subscriber, #39769) [Link]
Is UBIFS a journalling filesystem? Why? If you have random access storage
then there is no particular need to keep a journal all in one place. Rather than writing information synchronously to the journal you might as well write it synchronously to the real place it has to go.
First use of UBIFS? Posted Oct 10, 2008 16:04 UTC (Fri) by jamesh (subscriber, #1159) [Link]
From the look of the UBIFS web page, the data written to the journal becomes the new version of the data rather than being copied somewhere else when processing the journal.
I'd imagine that you'd want journalling of some form on flash for the same reason you'd want it on disk: to recover from power failures and crashes.
First use of UBIFS? Posted Oct 10, 2008 18:37 UTC (Fri) by dedekind (subscriber, #32521) [Link]
This is right. Our journal is "wandering". We accumulate data in the journal, and then commit, which means we just update the indexing information. Then we pick different eraseblocks for the new journal.
Also, the journal is multi-headed. One head is for data, one head is for stuff like inode met-adata, truncation nodes etc - this nodes are likely to become obsolete soon. One head is for Garbage collector. This way we minimize the amount of Garbage collection.
Potentially, it is possible to dynamically allocate journal heads and separate data ford different inodes. That would optimize things even more, but we did not implement this.
|
|
||||||||||||