I wouldn't make that statement just yet. ZFS still has some "issues", to say the least.
Sure, but they've got 6 months or so to sort them out and they've come a long way in the 8 months they approached Sun. As noted in the page I linked to, Apple seem to be releasing ZFS including features only weeks old in the main ZFS source so I'd imagine there's a fairly close relationship there with Sun.
ZFS is a virtual file system. You create 'pools' which can be spread over multiple devices on a network, RAID or other storage system. Remove one of the devices and it 'heals' itself.
ZFS is a virtual file system. You create 'pools' which can be spread over multiple devices on a network, RAID or other storage system. Remove one of the devices and it 'heals' itself.
Cool but is there any way it would actually be implemented in the mainstream PC desktop home/ small-to-big-enterprise markets? I'm afraid not. Not at least for 2007. As is the case with stuff that is really cool and smart, that doesn't make it to the big time.
Cool but is there any way it would actually be implemented in the mainstream PC desktop home/ small-to-big-enterprise markets? I'm afraid not. Not at least for 2007. As is the case with stuff that is really cool and smart, that doesn't make it to the big time.
I guess it depends on the GUI Apple shoves around it. Not many people are going to drop to Terminal and run zpool. Even then though it looks like there's still a few things to work out like boot support and it seems the Linux people are up in arms over the Sun/GPL licence incompatibility.
But, if Apple are sticking it in DiskUtility now, it'd be unlikely they've slipped in support by mistake and are taking it out again before Leopard ships later in 2007. Maybe it's an OSX server thing perhaps not workstation so maybe you're right - not quite mainstream. But it's promising.
I'd love it for webhosting - completely transactional self healing file system. Fantastic.
I think for laptops [MacBook particularly since it is so easy] replacing the hard disk with a Seagate Momentus 7200rpm SATA is the first step that will make s significant difference.
For desktops, I would say the Intel Matrix raid on Intel chipsets is so far the best mainstream enthusiast/ prosumer/ pro desktop solution. 4 SATA hard drives: Critical apps fast and protected, with a good RAID 0 section for speedy scratch disk/ temp workfile space/ game texture and level loading/ pagefile(s) etc. Image:
Raid 5 and Raid 0 on software raid That will eat up a lot of cpu time.
I thought, if I get a 1GB 1.83ghz MacBook White, I can easily swap out the 5400rpm drive for a 7200rpm Seagate Momentus. That would bring it to the cost of the 1GB 2ghz MacBook White, but tells ya what, I think a faster near-desktop-speed hard disk is much more worth the cost than 370mhz more in CPU power. Also given how overclockable the Core2 is in general (Conroes 2ghz can hit 3ghz easy on stock air cooling), I'm reluctant to pay that extra for a 1.83ghz Merom when that's just what it's binned at for power/watt/etc; it could easily be clocked at 2.00ghz at the expense of slightly higher heat and power draw. (Edit: Of course, there's no way to overclock the 1.83ghz in the MacBook to 2.00ghz, I'm happy with a 1.83ghz Merom with a 7200rpm 80gb HDD, 1GB RAM.. That would be a nice teh snappy all rounder. BUT I DON'T LIKE THE MACBOOK SCREEN - Too small, colour/contrast changes a bit *too* much when you shift viewing angle).
The performance tests of HDDs are bit tricky. These are often done when the HDD is relatively empty. Bare Feats had some interesting test results that you may want to consider. The results of the larger, 4200RPM drives were especially important to me as I tend max out my drives fairly quickly.
Quote:
CONCLUSIONS
These latest test results reveal that,
a) The 7200rpm drives are the fastest when empty...
b)... but if you have 74GB of data on each of the drives, the 4200rpm drive was actually faster. That's because 74GB of data puts the 7200rpm drive at 80% capacity while the 4200rpm drive is only at 40% capacity. (See "74G Mark" graphs above.)
The performance tests of HDDs are bit tricky. These are often done when the HDD is relatively empty. Bare Feats had some interesting test results that you may want to consider. The results of the larger, 4200RPM drives were especially important to me as I tend max out my drives fairly quickly.
I didn't know that, that's good to know, I'll have to read the page carefully to see if that's random access or just all out large file speed. Some sites show a graph where they show the read or write speed at a particular point on the drive, this is where it would be useful.
Your URL got messed up, here is the fixed version:
That assumes that the Matrix raid is CPU bound. Besides, I've never seen software RAID take any more than a negligible amount of CPU power.
Yeah Jeff I'm with you on this one, the Intel Matrix RAID and even nVidia on-board RAID on AMD mobos, I think have minimal CPU impact. AFAIK. Given that they are supposed to be pretty much hardware RAID implementations not software.
Yeah Jeff I'm with you on this one, the Intel Matrix RAID and even nVidia on-board RAID on AMD mobos, I think have minimal CPU impact. AFAIK. Given that they are supposed to be pretty much hardware RAID implementations not software.
You might have a bit of a point here. For striping, I don't think it takes any more either way. Software mirroring doubles the IO over hardware, but I've had software mirroring run a little faster than hardware mirroring. Software RAID-5 may take 20-25% more overall system IO than a hardware RAID-5 because the parity data goes down the entire path from CPU to drive rather than being generated at the hard drive controller. I'd need more information. For one, how much of the work the Matrix RAID does. One possibility that may make it a non-issue is that hard drives aren't terribly fast so it may or may not be a problem.
You might have a bit of a point here. For striping, I don't think it takes any more either way. Software mirroring doubles the IO over hardware, but I've had software mirroring run a little faster than hardware mirroring. Software RAID-5 may take 20-25% more overall system IO than a hardware RAID-5 because the parity data goes down the entire path from CPU to drive rather than being generated at the hard drive controller. I'd need more information. For one, how much of the work the Matrix RAID does. One possibility that may make it a non-issue is that hard drives aren't terribly fast so it may or may not be a problem.
Thanks to a little boost from Xmas I'm tempted to go build a Core2Duo [Conroe] XP2pro system and maybe try out the Matrix RAID too..... Or maybe frack all that I'll get a 17" iMac... Or maybe a 2nd hand Mac Mini because they will update the Mac Minis to Core2Duo in 4 months max.
Damn. Inspired to run some nice benchies for y'all. But there's quite a few for hard disks and it's actually challenging - how much is synthetic, what's "real", dependence on apparently how much each hard disk is filled, etc, etc.
Hell, maybe I'll just get some Jaegermeister for New Year's Eve. ...And some hookers. Gotta have the hookers.
Comments
I wouldn't make that statement just yet. ZFS still has some "issues", to say the least.
Sure, but they've got 6 months or so to sort them out and they've come a long way in the 8 months they approached Sun. As noted in the page I linked to, Apple seem to be releasing ZFS including features only weeks old in the main ZFS source so I'd imagine there's a fairly close relationship there with Sun.
So you missed the ZFS support leak then?
http://themachackers.com/2006/12/19/...a-closer-look/
RAID is about to become obsolete, at least as far as users at the front end are concerned.
It doesn't matter for most people either way.
I'm not finding much information there or anywhere on redundancy or failover with regard to ZFS. If it does have that, then that would be nice.
It doesn't matter for most people either way.
I'm not finding much information there or anywhere on redundancy or failover with regard to ZFS. If it does have that, then that would be nice.
Siracusa on Ars in his journal had a load of speculation on ZFS back before Apple were doing anything with it.
See http://en.wikipedia.org/wiki/Zfs also which is nice and concise
ZFS is a virtual file system. You create 'pools' which can be spread over multiple devices on a network, RAID or other storage system. Remove one of the devices and it 'heals' itself.
Siracusa on Ars in his journal had a load of speculation on ZFS back before Apple were doing anything with it.
See http://en.wikipedia.org/wiki/Zfs also which is nice and concise
ZFS is a virtual file system. You create 'pools' which can be spread over multiple devices on a network, RAID or other storage system. Remove one of the devices and it 'heals' itself.
Cool but is there any way it would actually be implemented in the mainstream PC desktop home/ small-to-big-enterprise markets? I'm afraid not. Not at least for 2007.
Cool but is there any way it would actually be implemented in the mainstream PC desktop home/ small-to-big-enterprise markets? I'm afraid not. Not at least for 2007.
I guess it depends on the GUI Apple shoves around it. Not many people are going to drop to Terminal and run zpool. Even then though it looks like there's still a few things to work out like boot support and it seems the Linux people are up in arms over the Sun/GPL licence incompatibility.
But, if Apple are sticking it in DiskUtility now, it'd be unlikely they've slipped in support by mistake and are taking it out again before Leopard ships later in 2007. Maybe it's an OSX server thing perhaps not workstation so maybe you're right - not quite mainstream. But it's promising.
I'd love it for webhosting - completely transactional self healing file system. Fantastic.
I think for laptops [MacBook particularly since it is so easy] replacing the hard disk with a Seagate Momentus 7200rpm SATA is the first step that will make s significant difference.
For desktops, I would say the Intel Matrix raid on Intel chipsets is so far the best mainstream enthusiast/ prosumer/ pro desktop solution. 4 SATA hard drives: Critical apps fast and protected, with a good RAID 0 section for speedy scratch disk/ temp workfile space/ game texture and level loading/ pagefile(s) etc. Image:
Raid 5 and Raid 0 on software raid That will eat up a lot of cpu time.
Raid 5 and Raid 0 on software raid That will eat up a lot of cpu time.
That assumes that the Matrix raid is CPU bound. Besides, I've never seen software RAID take any more than a negligible amount of CPU power.
I thought, if I get a 1GB 1.83ghz MacBook White, I can easily swap out the 5400rpm drive for a 7200rpm Seagate Momentus. That would bring it to the cost of the 1GB 2ghz MacBook White, but tells ya what, I think a faster near-desktop-speed hard disk is much more worth the cost than 370mhz more in CPU power. Also given how overclockable the Core2 is in general (Conroes 2ghz can hit 3ghz easy on stock air cooling), I'm reluctant to pay that extra for a 1.83ghz Merom when that's just what it's binned at for power/watt/etc; it could easily be clocked at 2.00ghz at the expense of slightly higher heat and power draw. (Edit: Of course, there's no way to overclock the 1.83ghz in the MacBook to 2.00ghz, I'm happy with a 1.83ghz Merom with a 7200rpm 80gb HDD, 1GB RAM.. That would be a nice teh snappy all rounder. BUT I DON'T LIKE THE MACBOOK SCREEN
The performance tests of HDDs are bit tricky. These are often done when the HDD is relatively empty. Bare Feats had some interesting test results that you may want to consider. The results of the larger, 4200RPM drives were especially important to me as I tend max out my drives fairly quickly.
CONCLUSIONS
These latest test results reveal that,
a) The 7200rpm drives are the fastest when empty...
b)... but if you have 74GB of data on each of the drives, the 4200rpm drive was actually faster. That's because 74GB of data puts the 7200rpm drive at 80% capacity while the 4200rpm drive is only at 40% capacity. (See "74G Mark" graphs above.)
http://barefeats.com/mbcd7.html200GB
The performance tests of HDDs are bit tricky. These are often done when the HDD is relatively empty. Bare Feats had some interesting test results that you may want to consider. The results of the larger, 4200RPM drives were especially important to me as I tend max out my drives fairly quickly.
I didn't know that, that's good to know, I'll have to read the page carefully to see if that's random access or just all out large file speed. Some sites show a graph where they show the read or write speed at a particular point on the drive, this is where it would be useful.
Your URL got messed up, here is the fixed version:
http://barefeats.com/mbcd7.html
That assumes that the Matrix raid is CPU bound. Besides, I've never seen software RAID take any more than a negligible amount of CPU power.
Yeah Jeff I'm with you on this one, the Intel Matrix RAID and even nVidia on-board RAID on AMD mobos, I think have minimal CPU impact. AFAIK. Given that they are supposed to be pretty much hardware RAID implementations not software.
Yeah Jeff I'm with you on this one, the Intel Matrix RAID and even nVidia on-board RAID on AMD mobos, I think have minimal CPU impact. AFAIK. Given that they are supposed to be pretty much hardware RAID implementations not software.
But they may eat up of a lot of chip set io
But they may eat up of a lot of chip set io
Yeah... Possibly. I haven't seen/ investigated definitive benchmarks recently on the impact on CPU, Chipset I/O...
But they may eat up of a lot of chip set io
You might have a bit of a point here. For striping, I don't think it takes any more either way. Software mirroring doubles the IO over hardware, but I've had software mirroring run a little faster than hardware mirroring. Software RAID-5 may take 20-25% more overall system IO than a hardware RAID-5 because the parity data goes down the entire path from CPU to drive rather than being generated at the hard drive controller. I'd need more information. For one, how much of the work the Matrix RAID does. One possibility that may make it a non-issue is that hard drives aren't terribly fast so it may or may not be a problem.
You might have a bit of a point here. For striping, I don't think it takes any more either way. Software mirroring doubles the IO over hardware, but I've had software mirroring run a little faster than hardware mirroring. Software RAID-5 may take 20-25% more overall system IO than a hardware RAID-5 because the parity data goes down the entire path from CPU to drive rather than being generated at the hard drive controller. I'd need more information. For one, how much of the work the Matrix RAID does. One possibility that may make it a non-issue is that hard drives aren't terribly fast so it may or may not be a problem.
Thanks to a little boost from Xmas I'm tempted to go build a Core2Duo [Conroe] XP2pro system and maybe try out the Matrix RAID too.....
Damn. Inspired to run some nice benchies for y'all. But there's quite a few for hard disks and it's actually challenging - how much is synthetic, what's "real", dependence on apparently how much each hard disk is filled, etc, etc.
Hell, maybe I'll just get some Jaegermeister for New Year's Eve.
http://techreport.com/reviews/2005q1...d/index.x?pg=1