Page 1 of 1

fastest filesystem?

Posted: Fri May 16, 2008 7:31 pm
by darkpaw
Can anybody suggest a filesystem that has faster read/write than ext3?
Settomg noatime is a help slightly, but having higher performance would be nice. Ext3 isn't exactly known for its performance, even though with some tuning it does help (in fact, I think I even suggested this somewhere).

But what about other filesystems? Is anyone using Reiserfs? xfs?

Thoughts on other fs types?

Posted: Fri May 16, 2008 8:11 pm
by coke
I'm using reiserfs. I got the impression somewhere that it handled smaller files (like a bajillion jpegs) better than the alternatives. Not sure about speed, but it should give you more space.

Posted: Sat May 17, 2008 1:16 am
by cordel
I just but a system together with xfs. I should know in a month about how well it does.
I might put something together with several drives and mirror the content between xfs and reiser and se if I can benchmark it if I don't find anything on line, but I'm sure someone has benchmarked these and it's just a matter of STFW.
I know wikipedia has alot of information on the different FS's I just haven't dug around.

While burning the system in it recorded around 120GB and it took about 24hrs to remove with zmaudit after removing the events from the DB with the system recording 320x240 Color at 60 fps to that single drive, so no raid, no LVM, No spanning of any type.
Seagate Barracuda ES.2 ST3500320NS 500GB 7200 RPM 32MB Cache SATA 3.0Gb/s Hard Drive

speed reasons

Posted: Sat May 17, 2008 3:53 am
by darkpaw
My reason for speed is writing, in particular. I want to be able to dump as many jpgs as I can onto the drive without losing frame rate. Not just on personal installs, but on customer installs. Anything to improve performance is always welcome. :)

In my personal setup, I am running the following:

Pentium D dual-core (model 915, I think), 2.8 GHz overclocked (currently, I've been messing with speed ratios) to 3.2 GHz, 3 Gig RAM (DDR2-800, I think), 3 x 146 Gig U320 15k-RPM SCSI drives on a PERC3 RAID controller running hardware RAID-5. I realize a drawback of write speed with RAID-5 over RAID-0, but I want the drive redundancy in my installs so I try to go with something along these lines for any higher-end setup. Any system I install is running Fedora Core (8, in this case).

ext3 seems to do "OK" for writes, but if multiple cameras are writing at 15-20fps, there are some drops. CPU load never goes over 1.5 (2.0 on a dual-core is like running 1.0 on a single-core/CPU machine, working well but not over-working). Writing keeps up "for the most part", but as I mentioned, it gets some dropped frames occasionally. This system has 5 cameras all running 640x480 (analog bttv) with 3-4 zone modetect on each camera. When cameras are idle, it runs at .1-.2 load....when running full rate on multiple cameras at once, with a montage up, it pegs 1.5 on the load. The system runs as server only, run-level 3, with everything disabled except what it needs to run the web server and ZM. It also serves as my primary web server for outside ZM. I run extended ring there are LOTS of jpegs being dumped to the drive every interval...and I LOVE using gobs of shared memory. :)

I've been reading a bit recently about the hack for libjpeg apparently giving more speed on jpeg writes. Is there truth to this?

Also, anyone do any tweaking on hardware-based RAID cards? I'm using cached writes, trying to utilize the cache on the PERC controller. I could likely bump up the RAM in the PERC, if that might help a bit, as well. Or play with other coptions (copyback, I assume, would slow it down tremendously).

Posted: Mon Jun 02, 2008 5:25 pm
by ammaross
I have heard ext2 has better performance than ext3, and also that the new ext4 that Fedora 9 ships with has a significant performance improvement over ext3. I don't have any benchmarks however. I just merrily throw my jpegs at ext2 and hope for the best. :)

Oh, and ext4 solves the 32000 files in one folder issue.


Posted: Mon Jun 02, 2008 6:50 pm
by darkpaw
ext2 is dangerous. There is no journaling, so if there is a crash during a write, your filesystem is toast.

Will ZM 1.22.3 work in FC9 yet? What version of ZM is in the FC9 repo?

Posted: Mon Jun 02, 2008 6:59 pm
by ammaross
I don't use the ZM rpms as I have numerous compile mods as well as post-compile interface mods, so it is easier for me just to build from scratch. ZM works fine on FC9, and I believe there is a few posts on how to get it to work (ffmpeg-related problems and the like).

Yes, the journaling thing is a problem, but I'm not terribly concerned about it at the moment. I'd prefer the performance over journaling until I get the system updated to ext4.

Also, I'd frankly like to use 1.23.3 over 1.22.3, but my PTZ script only works in 1.22.x due to the perl-only PTZ interface in 1.23.x :( I miss the DVR functionality.