I've been looking at various Linux filesystem benchmarks and it's clear that some filesystems perform better then others. But aside from the benchmarks, I was wondering if the difference is really noticable in real-world scenarios? For example, would I *really notice* a faster boot-up time? And what about performance when transfering files over a network using NFS? I am looking at filesystems as a way to boost performance on my aging Pentium 2 specification laptop. From what I have read, JFS seems the best pick in this scenario because of it's low CPU requirements. However it seems few people use JFS and I am wondering how well it's supported? I am using Xubuntu Feisty Fawn on the laptop.
JFS is a solid and well-supported filesystem. You shouldn't have any problems with it. I only recommend that if you want to use a non-ext filesystem, you should at least put your /boot partition on ext2 or ext3 in order to keep your bootloader as simple as possible. I use JFS or XFS for most things, but ext3 still has its uses. If you use an ext3 filesystem and want more performance, you can do so with an old Linux sysadmin trick, by adding the 'noatime' flag to the mounted ext3 filesystem(s) in your /etc/fstab, like so: Code: # Before tweak: LABEL=/ / ext3 defaults 1 1 # After tweak: LABEL=/ / ext3 noatime 1 1 That will remove the 'last access time' write that normally occurs when you read a file, basically gaining you a free performance boost. It usually doesn't matter when you've read a file, only when you've touched it. The only occasion where the noatime tweak should be avoided is on partitions hosting news spoolers. So unless you're running listserv at home (extremely unlikely), you can safely use noatime. P.S. - The noatime tweak should work on most any Linux/UNIX filesystem, not just ext3
i've only used ext2 & ext3 but my last few OSs were on ext3, i knew there were hundreds of FSs but i thought they were only for the OS they were running BTW: how do you format a partition to, say: JFS, and what are the advantages (performance or whatever)[ot]just compiled my first prog from source[/ot]BTW: cant beleive ive been using *nix for over a year without compiling from source[ot]pwease be kind[/ot]
The laptop's hard drive is quite small, only 6.5GB, so I tend to make a partition for root and a swap partition. I decided to wipe the old Xubuntu 7.04 installation and start a fresh, this time formatting the whole root partition as JFS. I was a little concerned about GRUB's ability to boot a JFS partition, but I needn't have worried! It works flawlessly! It's difficult to do a direct comparision with the old installation because the old installation wasn't a fresh install. Rather, it had previously upgraded Edgy Eft, which had upgraded Dapper Drake and then even Breezy Badger! So of course a fresh installation, in theory, should provide better results then an upgrade. Anyhows, the boot up time was comparible, if ever so slightly better. But the real difference was in how much faster the applications load! I click on Mousepad and boom, it's up before you know it! This is pretty amazing considering we are talking about a 333Mhz laptop here! I've also noticed less hard drive activity and slightly lower system resources. I've now determined that JFS is a viable alternative to EXT, especially on an older computer. However it's longevity is a concern, what with EXT4 and Reiser4 (if it ever gets in the kernel) just around the corner. You can specify JFS at installation, use Gparted or use the command line [ot] Cool man! Well done! What did you compile? The first program I ever compiled is Gens, a Sega emulator. It wasn't the easiest choice because it requires an earlier GCC, so I had to change some environment variables. I then learned how to produce debian packages and decided to package Gens! I host my Gens package on the Ubuntu Forums. I've since compiled and packaged Generator, which is another Sega Genesis emulator.[/ot]
AT, have you used noatime with UFS or UFS2? I know that it's an available option but I've never used it. Does it help FS performance in that case? Any other performance tweeks for UFS2 I should know about? Also, would noatime help performance of a USB external hard disc formatted UFS2 (used for rsync backups only)? How about any freeBSD or Linux swap performance tweeks? Thanks
Yes, and yes. Eliminating an arbitrary write operation speeds up your read performance noticeably, and in turn will give you a write boost as well since you're freeing up the HDD's physical IO a bit. Never had a problem with noatime on UFS*, with or without softwrites enabled. However, as I stated before, you want to avoid using noatime on any partition which will house a listserv. Linux and FreeBSD already use swap as a matter of last resort only, or to clean "dirty" writeback from RAM. In other words, even if there was a tweak, you wouldn't notice any difference unless you didn't have nearly enough RAM for your system, since UNIX systems use RAM in an extremely efficient manner. That said, swap is designed as a special format type specifically for this reason; it's already optimized especially for swaping pagefiles.
Thanks! I will give it a go. I was wondering about swap performance because my girlfriend's computer has only 128 MB of RAM and uses swap just about all the time. The system is running MEPIS 6.5. I trimmed the 'fat' a bit but it still need swap space for most operations. I need to optimize swap effeciancy and general disk IO speed. Someday she will get a RAM upgrade but for now its a little slow. It was a free computer so I'm not complaining.
i still struggle with with konsole, i prefer a GUI (but, *nix seems to be heading that way) Gparted only has available FAT16, FAT32, Ext2, Ext3, swap & resiser all other formats are grayed out[ot] thankies GNU parted but i prefer GUI mode WOW WOW slow down please[/ot]BTW: i think my router or integrated NIC may be on its way out, cos yesterday i couldn't get online & i usually have to update Autherisation & DNSs[ot]i think its the router[/ot]might order a router & NIC while net connection working, NIC cheaper so ill try that first[ot]wasn't working with the exact same filter & ethernet cable that IS working now[/ot]
