What is Digg?81 Comments
- db113456, on 10/10/2007, -1/+18Very interesting benchmark, specially with ext4 being in (experimental) and (testing) mode :-)
Still very interesting , really. - lolwtfhaha, on 10/10/2007, -0/+16Useless... not repeatable. Where are the exact switches used for each benchmark? All the ">99999" for bonnie means nothing-- the file sample size was too small to generate meaningful numbers. Document exact procedures benchmarkers! :-)
- cfd339, on 10/10/2007, -5/+17What, no FAT32 in that test? ;-)
- trogdoor, on 10/10/2007, -4/+15He actually had written that part of the article but the file got corrupted.
- inactive, on 10/10/2007, -0/+10Nope, I was genuinely interested in the various comparisons of the next-gen formats. If Ext4 keeps up the ratings, I believe that I shall be going with it once I reformat and when the recovery software catches up.
- theendlessnow, on 10/10/2007, -0/+10reiserfs v3 is the only filesystem that comes close to ZFS in features. XFS lacks data journaling, ext3/4 lack decent dynamic growing options (and have many limitations besides that). Leaving reiserfs out is a mistake IMHO. With that said, with reasonable hardware, you can pretty much saturate things to where the bandwidth to the disk is the primary factor.
See: http://www.ntlug.org/Presentations/FilesystemBenchmarks - annenk38, on 10/10/2007, -0/+10whatever may be the case with Hans Reiser, reiserfs v3 is stable and feature-complete.
- 2kude, on 10/10/2007, -2/+11How about a visualization?
- neomis, on 10/10/2007, -2/+11I guess my question is where is reiserfs or reiser4 in the comparison?
- Guspaz, on 10/10/2007, -1/+10It's stupid to benchmark ZFS and not include a comparison of RAIDZ or RAIDZ2. And for those who want to jump in and say that RAIDZ is less reliable, might I point out that with 4 drives (the minimum for RAID10), RAIDZ2 has the same redundancy, you can lose any two drives.
RAIDZ is similar to RAID5, but with some improvements for reliability and performance (no write hole, per-block checksums in addition to parity, parity is distributed, dynamic stripe sizes, etc.), and RAIDZ2 simply adds another set of parity bits, meaning there are effectively three "copies" of any given block.
I'm not trying to sound like an advertisement, it's just that RAIDZ is one of ZFS's biggest features, and to benchmark RAID on ZFS without it doesn't make any sense. - inactive, on 10/10/2007, -2/+10You shouldn't be on digg.
- Genady, on 10/10/2007, -0/+7Meh. It's hard to tell if the author groks that ZFS is more than a file system perse, but more of a LVM with file system thrown in. It's not surprising to me at all that XFS is quick on sequential things, media manipulation was SGI's bread and butter when XFS grew up, on RAID 4 sets. You've probably got some oddness in there too because of different Kernel structures between Linux and Solaris. Maybe someone will re-run the tests in BSD (Does EXT4 run on BSD?)
- jambarama, on 10/10/2007, -0/+6The thing that really surprised me was ZFS. XFS and EXT4 had pretty dang good performance almost throughout (almost). ZFS was way the heck all over the place, apparently it blows* reading software RAIDs had does ok with hardware RAIDs. Generally the performance isn't great (maybe not such a great SAN file system) compared to EXT4 and XFS, but you can't beat the features! Now I wonder how the new Reiser would fare... I'd be willing to test if he'd hand over his hardware!
*Blows in this case means - runs faster than anything I've ever run, but slow as heck compared to competitors. - db113456, on 10/10/2007, -2/+7Excel is non-free software, use Open Office Spreadsheet or Gnumeric or Kspread instead.
- IHatePants, on 10/10/2007, -0/+5I can't really say much about performance between Solaris and Linux since I mainly run J2EE apps on the two of them (JVM implementations can distort benchmarks), but when it comes to stability, I've seen little difference. My Linux boxes only need restarts when I upgrade the kernel. I'm not saying that Linux is more stable than Solaris, but Linux is as stable as I'll ever need a server OS to be as a J2EE server.
- Benad, on 10/10/2007, -2/+6ZFS does a checksum on every block, so that's why overall throughput is lower than ext4. And of course copies take no time whatsoever (copy-on-write).
- rootstyle, on 10/10/2007, -1/+5...and this is why digg has gone down hill... What started as a nice alternative to /.'s retarded editors, is now the youtube of news sites... Yay for more chain mail pics from the mid 90s on the front page :p
On a less negative note, great round up, I've always been a fan of XFS (in the IRIX days it was just mind blowing compared to EFS :) but great stuff coming from ext4 though ! That will definitely be one project to watch, and I'm sure will be the defacto for new distros in a couple years time. - MeneerR, on 10/10/2007, -0/+4It wasn't about Ubuntu until YOU brought it up. See what YOU DID THERE?
You are a troll and need to be ass-raped by a prehistoric monkey - grumpyrain, on 10/10/2007, -0/+4How about a gif or png image of a graph generated by whatever spreadsheet program your philosophy permits you to use?
(And how about digg gets this session expired bug fixed!) - zonk3r, on 10/10/2007, -0/+3Testing these new filesystems is sort of like testing a piece of software at a 0.7 version number where things are still being implemented and optimized. Sun for example will be the first to tell you that ZFS isn't complete yet. Hell I know some people that only recently moved to EXT3 because they needed to be damned sure of stability and they didn't trust a "new" FS. Most filesystems take a few years of work to get to a point where they are stable, fast and feature laden.
- abhiroop, on 10/10/2007, -1/+4Nope I opened it because I wanted to figure out which would be a good format to use in my next debian install.
- DiggerT, on 10/10/2007, -0/+3and the overall winner in the benchmark is?
- digguserer, on 10/10/2007, -0/+3And if they had turned off the checksums and used a marvell non-raid card, zfs would have run away with most of the numbers in the test. Some of the tests that were done were highly prone to differences across platforms, so I would take everything here with a grain of salt. Bottom line is that if you want to find out what works best for your workload, test it yourself. In many cases, the FS is the least of your performance concerns and you should really be looking at features and reliability.
- jambarama, on 10/10/2007, -0/+3Bad form replying to myself - but I wonder how much of the performance differential is just Solaris v. Linux. I've read Linux has the edge in performance, though Solaris certainly has the edge in stability.
- divergentdave, on 10/10/2007, -1/+4I was annoyed that the author tested ZFS using hardware and software RAID systems, considering that ZFS is supposed replace traditional RAID systems through mirroring in the zpool. Let ZFS itself take care of redundancy, not some expensive RAID controller.
- bnolsen, on 10/10/2007, -0/+3Pretty much reiserfs is the only filesystem I've used with long term success on extremely large, busy hardware raid systems (ie, in excess of 10TB).
XFS is scary just because the amount of code in the driver is huge, and additionally its missing data journalling.
JFS seems to be a good option but seems to "wig out" within a couple of weeks during testing. Been this way for a few years.
ZFS's performance numbers,especially sequential, are interesting. Solaris may well extremely poor hardware storage drivers and perhaps very poor IO scheduling and low level block handing. Considering ZFS is supposed to take care of everything from disk to volume management this is pretty discouraging and makes me think ZFS is still nothing more than a BETA filesystem. - noksagt, on 10/10/2007, -0/+2Anecdotal experience only, but we had a power outage that took down servers with different file systems. We had to recover a reiser4 partition. It lost some file names, but we had all data. We also had two XFS partitions go down & could recover neither. I think both XFS and reiser4 are maintenance head aches. I actually like reiser4, but too many recent patch sets have had corruption if the CPU load was too high. It would've been nice to see it cleaned up, stabilized, and mainlined.
I like JFS. It is a godsend for my laptop--power usage is TINY!
But we, like bnolsen, will stick with reiserfs (reiser3 has had none of the stability issues described for reiser4) and/or the ext series in future systems (heck--we have plenty of boxes that can only run UFS, so reliability wins out over performance and "features."
I've recovered a hosed reiser4 (power ofutage) that - chmoder, on 10/10/2007, -1/+3Coooelll. Digg it! Good Job. I wish There was a clear winner. not an "Im better than you at this but not this." I guess the winner is whoever is faster for what you need.
- geniusj, on 10/10/2007, -0/+2raidz2 (similar to RAID-6) has even more redundancy than a RAID-10 of 4 drives, because, as you mentioned you can lose ANY two drives. Lose the wrong two in a RAID-10 and you're screwed.
- gyllstromk, on 10/10/2007, -1/+3"This is my attempt to cut through the hype and uncertainty to find a storage subsystem that works."
This article should really emphasize that it is limiting evaluation to purely a transfer speed test (and not a very thorough one at that). Of course, there are many other important qualities of a file system that should be assessed before determining which one "works".
One could argue that a journaled file system is slower than one without a journal. Yes, ZFS might be slower, but it also has transactional file operations (screw journals), provides constant time snapshotting, incremental backups, file system cloning, a simple pooling system, compression (which can speed the file system up in exchange for using the CPU), and a very robust RAID. Climb out of your LOONIX hole. - K3ITHK, on 10/10/2007, -0/+2I was looking for FAT16.
- inactive, on 08/26/2008, -1/+3Very interesting... Yes, quite... I definitely understood every word.. Indeed...
- Tephra, on 10/10/2007, -1/+3lets be clear here - this ISN'T a review of the filesystems against each other .... its purely performance based.
zfs beats both ext and xfs in terms of data reliability...
also ext sux for deletion of BIG files (gigabytes - ie tv recordings) but xfs is very good for that specific task - skyshock1, on 10/10/2007, -2/+4Which one is optimized for solid state storage? THAT's the one I would go with. HDD w/ spinning platters will be a thing of the past VERY soon.
- abhiroop, on 10/10/2007, -0/+2wait when u say "on the other hand for Linux Ext3/4 is the best all around " what does that mean? Is ext3 or ext4 better? I'm interested in knowing because Ubuntu use ext3, but if ext4 is relatively stable and faster I wouldn't mind using it.
- gyrfalcon, on 10/10/2007, -0/+2Thanks for pointing out half the problem with digg :)
- coszmos, on 10/10/2007, -1/+3Meh. Try again with a jbod using zfs vs. lvm2 with either software or hardware raid. Also extrapolate the costs of administrative overhead in managing the gear when something (simulated) fails. Now that would be an interesting comparison, something that would be useful to the person managing this crap; ie me.
- maninalift, on 10/10/2007, -0/+1Its not like speed_of_beta = 0.7 * speed_of_release.
Just beta may not be a true reflection of the refined product one may improve and another may not. - sirmo, on 10/10/2007, -1/+2I really hoped that flash could bring us to mainstream solid state drives but. Flash still has many limitations other than being more expensive. Most notable limitation is that flash memory has a finite number of erase-write cycles.
Battery-backed static RAM on the other hand relies on batteries.
Other than being expensive both types of solid state technologies have some major drawbacks. I am afraid we will have to stick with conventional platter hard disks for awhile. - Nelson69, on 10/10/2007, -0/+1You'd be hard pressed to show some actual data that showed ZFS being more reliable than either.
- digguserer, on 10/10/2007, -0/+1It is also blazing fast for most (not all) workloads. There are even performance reports out there comparing it to veritas and showing zfs edging it out in most categories. Also, if you don't need checksumming, disabling that will also speed it up. Most performance problems for zfs come down to the drivers and controller used. Get those right and it will not let you down.
- gclef, on 10/10/2007, -0/+1That's only the case if your Java app is coded to use threads properly. For example, I ran into a situation where a cross-platform Java app (from a vendor, so I couldn't just fix it) worked *horribly* on Solaris, mostly because the app used at most 2 threads. Since SPARC CPUs (which we were using at the time) have slower clock speeds than the x86 ones, the performance of this app was much better on x86.
Sun may optimize the JRE for Solaris, but it requires the app author to actually design the app to take advantage of that. In my experience, they don't. - starheart, on 10/10/2007, -0/+1The author isn't clear if it is single core or dual core. Software raid would greatly depend on how many cores the system has. If it was just single core, then the software raid results could be a lot better by just making it a dual or quad core system.
- geniusj, on 10/10/2007, -0/+1ZFS will be slower than most traditional filesystems currently. But it's a fairly new ground-up effort compared to the others (ext4 is based on ext3 I assume?). It's really everything else about ZFS that rocks currently. It gives you a ton of flexibility and is very easy to manage. Not to mention the reliability as you mentioned.
- MeneerR, on 10/10/2007, -1/+264 kb should be enough for everyone.
Dude, solid state will always be more $$/GB.
We will always want more space.
The day we can buy 120 gb flash drives, is the day we can buy 3 tb hd for the exact same price.
Guess which one can hold more than 100 hd-movies and which one can't.
So, to answer your question. In the future, you will have BOTH.
You will have solid-state (about 6gb maximum) that actually holds the system and the swap file/partition.
You will have a spinning magnetic harddrive that is holding all your movies and game-data. - GMorgan, on 10/10/2007, -0/+1Java should perform better on Solaris though, Sun go the extra mile using kernel threads for Solaris while the Java implementation for Linux only uses software threads (as it does for Windows) AFAIK.
-
Show 51 - 81 of 81 discussions
The Digg Toolbar for Firefox lets you Digg, submit content, and keep track of Digg even when you're not on the Digg site. Download the official 
© Digg Inc. 2009
— Content created and posted by Digg users is