Messages in this thread | | | Date | Wed, 26 May 1999 02:01:46 -0400 (EDT) | From | "Theodore Y. Ts'o" <> | Subject | Re: 2.3.x wish list? |
| |
Date: Mon, 24 May 1999 17:57:11 +0200 (MET DST) From: marcel@mesa.nl (Marcel J.E. Mol)
If it is root that is writing, it will use the reserved space anyway so depending on the reserved size it will take longer or shorter for root to fill the fs whiel others can't write to it. There might be fs that will not have any root activity, so a root cushion is not needed. If for some reason you need emergency space you can always borrow from other fs.
If you read my original message, you will see that I said this was mostly true for the /usr and /var partitition, where root does need to write log files, so it's important that root have a cushion.
> The second > reason is that ext2's fragmentation-avoidance algorithms work best when > there a certain amount of slack space reserved in the filesystem (i.e., > we want to leave some amount of choice for the block allocation > algorithsm to choose optimal blocks to avoid fragmentation). That's why > -m is represented as a percentage, and why 5% is used as the default.
If the reserved space is used for allocations then in the end also the reserved space will be fragmented. In that case it all depends on the dynamics on the fs. But sooner or later you will need to defragment in order to improve fs performance and restore the contigious reserved space.
It depends on what how big the files you are writing. It's not necessary for the free space to be contiguous; just large enough so that files written to the disk can still be contiguous. For example, for an 5% of 8 gig is 40 megs. If most of the files you write are smaller than 1 meg, it's likely that the files can be stored contiguously, even if the filesystem only has 5-10% free space on it.
It is not that I question the use of reserved space. I just want to have finer control over it. In the end it is just a tuning parameter. The default can still be 5%. Allowing fraction percentages might also do the job...
That's reasonable. The next time I mess with tune2fs and mke2fs, I'll see what I can do on that front.
- Ted
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |