Re: to gmirror or to ZFS
But then zfs doesn't access every block on the disk does it, only the allocated ones On 20 July 2013 21:07, Daniel Feenberg feenb...@nber.org wrote: On Sat, 20 Jul 2013, Steve O'Hara-Smith wrote: On Sat, 20 Jul 2013 18:14:20 +0100 Frank Leonhardt fra...@fjl.co.uk wrote: It's worth noting, as a warning for anyone who hasn't been there, that the number of times a second drive in a RAID system fails during a rebuild is higher than would be expected. During a rebuild the remaining drives get thrashed, hot, and if they're on the edge, that's when they're going to go. And at the most inconvenient time. Okay - obvious when you think about it, but this tends to be too late. Having the cabinet stuffed full of nominally identical drives bought at the same time from the same supplier tends to add to the probability that more than one drive is on the edge when one goes. It's a pity there are now only two manufacturers of spinning rust. Often this is presummed to be the reason for double failures close in time, also common mode failures such as environment, a defective power supply or excess voltage can be blamed. I have to think that the most common cause for a second failure soon after the first is that a failed drive often isn't detected until a particular sector is read or written. Since the resilvering reads and writes every sector on multiple disks, including unused sectors, it can detect latent problems that may have existed since the drive was new but which haven't been used for data yet, or have gone bad since the last write, but haven't been read since. The ZFS scrub processes only sectors with data, so it provides only partial protection against double failures. Daniel Feenberg NBER -- Steve O'Hara-Smith st...@sohara.org __**_ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/**mailman/listinfo/freebsd-**questionshttp://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-** unsubscr...@freebsd.org freebsd-questions-unsubscr...@freebsd.org __**_ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/**mailman/listinfo/freebsd-**questionshttp://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-** unsubscr...@freebsd.org freebsd-questions-unsubscr...@freebsd.org ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On 21/07/2013 17:31, Steve O'Hara-Smith wrote: On Sun, 21 Jul 2013 14:13:39 +0930 Shane Ambler free...@shaneware.biz wrote: On 21/07/2013 04:42, Steve O'Hara-Smith wrote: It's a pity there are now only two manufacturers of spinning rust. I thought there was three left - Seagate WD and Toshiba I assumed Toshiba were out of the game, I've never seen anything bigger than 500GB with a Toshiba label. I have a 2.5 1TB Toshiba USB drive here. I see Toshiba 2 and 3TB 3.5 listed online. As I recall the Hitachi selloff - WD got the 2.5 Toshiba got the 3.5 I think the split was the only way to get the takeover approved. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On Sun, 21 Jul 2013 14:13:39 +0930 Shane Ambler free...@shaneware.biz wrote: On 21/07/2013 04:42, Steve O'Hara-Smith wrote: It's a pity there are now only two manufacturers of spinning rust. I thought there was three left - Seagate WD and Toshiba I assumed Toshiba were out of the game, I've never seen anything bigger than 500GB with a Toshiba label. -- Steve O'Hara-Smith | Directable Mirror Arrays C:WIN | A better way to focus the sun The computer obeys and wins.|licences available see You lose and Bill collects. |http://www.sohara.org/ ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
Steve O'Hara-Smith st...@sohara.org wrote: It's a pity there are now only two manufacturers of spinning rust. I didn't think there were _any_! Haven't oxide-coated platters gone the way of the dodo bird? ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On Sun, 21 Jul 2013 00:27:01 -0700 per...@pluto.rain.com (Perry Hutchison) wrote: Steve O'Hara-Smith st...@sohara.org wrote: It's a pity there are now only two manufacturers of spinning rust. I didn't think there were _any_! Haven't oxide-coated platters gone the way of the dodo bird? Ah the technicalities, this is a software group :-) -- Steve O'Hara-Smith | Directable Mirror Arrays C:WIN | A better way to focus the sun The computer obeys and wins.|licences available see You lose and Bill collects. |http://www.sohara.org/ ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On 16/07/2013 20:48, Charles Swiger wrote: Hi-- On Jul 16, 2013, at 11:27 AM, Johan Hendriks joh.hendr...@gmail.com wrote: Well, don't do that. :-) When the server reboots because of a powerfailure at night, then it boots. Then it starts to rebuild the mirror on its own, and later the fsck kicks in. Not much i can do about it. Maybe i should have done it without the automatic attachment for a new device. It's normally the case that getting a hot spare automatically attached should be fine, but not if you also have the box go down entirely and need to fsck. I'm more used to needing to explicitly physically swap out a failed mirror component, in which case one can make sure the system is OK before the replacement drive goes in. Agreed. Blaming gmirror for this kind of thing overlooks the overall design and operating procedures of the system, and assuming ZFS would have been any better may be wishful thinking. I've had plenty of gmirror crashes over the years, and they have all been recoverable. One thing I never allow it to do is to rebuild automatically. That's something for a human to initiate once the problem has been identified, and if it's flaky power in the data centre the job is postponed until I'm satisfied it's not going to drop during the rebuild. IME, one power failure is normally followed by several more. It's worth noting, as a warning for anyone who hasn't been there, that the number of times a second drive in a RAID system fails during a rebuild is higher than would be expected. During a rebuild the remaining drives get thrashed, hot, and if they're on the edge, that's when they're going to go. And at the most inconvenient time. Okay - obvious when you think about it, but this tends to be too late. Regards, Frank. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On Sat, 20 Jul 2013 18:14:20 +0100 Frank Leonhardt fra...@fjl.co.uk wrote: It's worth noting, as a warning for anyone who hasn't been there, that the number of times a second drive in a RAID system fails during a rebuild is higher than would be expected. During a rebuild the remaining drives get thrashed, hot, and if they're on the edge, that's when they're going to go. And at the most inconvenient time. Okay - obvious when you think about it, but this tends to be too late. Having the cabinet stuffed full of nominally identical drives bought at the same time from the same supplier tends to add to the probability that more than one drive is on the edge when one goes. It's a pity there are now only two manufacturers of spinning rust. -- Steve O'Hara-Smith st...@sohara.org ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On Sat, 20 Jul 2013, Steve O'Hara-Smith wrote: On Sat, 20 Jul 2013 18:14:20 +0100 Frank Leonhardt fra...@fjl.co.uk wrote: It's worth noting, as a warning for anyone who hasn't been there, that the number of times a second drive in a RAID system fails during a rebuild is higher than would be expected. During a rebuild the remaining drives get thrashed, hot, and if they're on the edge, that's when they're going to go. And at the most inconvenient time. Okay - obvious when you think about it, but this tends to be too late. Having the cabinet stuffed full of nominally identical drives bought at the same time from the same supplier tends to add to the probability that more than one drive is on the edge when one goes. It's a pity there are now only two manufacturers of spinning rust. Often this is presummed to be the reason for double failures close in time, also common mode failures such as environment, a defective power supply or excess voltage can be blamed. I have to think that the most common cause for a second failure soon after the first is that a failed drive often isn't detected until a particular sector is read or written. Since the resilvering reads and writes every sector on multiple disks, including unused sectors, it can detect latent problems that may have existed since the drive was new but which haven't been used for data yet, or have gone bad since the last write, but haven't been read since. The ZFS scrub processes only sectors with data, so it provides only partial protection against double failures. Daniel Feenberg NBER -- Steve O'Hara-Smith st...@sohara.org ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On 21/07/2013 04:42, Steve O'Hara-Smith wrote: It's a pity there are now only two manufacturers of spinning rust. I thought there was three left - Seagate WD and Toshiba ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On Jul 16, 2013, at 11:42 AM, Warren Block wrote: On Tue, 16 Jul 2013, aurfalien wrote: On Jul 16, 2013, at 2:41 AM, Shane Ambler wrote: I doubt that you would save any ram having the os on a non-zfs drive as you will already be using zfs chances are that non-zfs drives would only increase ram usage by adding a second cache. zfs uses it's own cache system and isn't going to share it's cache with other system managed drives. I'm not actually certain if the system cache still sits above zfs cache or not, I think I read it bypasses the traditional drive cache. For zfs cache you can set the max usage by adjusting vfs.zfs.arc_max that is a system wide setting and isn't going to increase if you have two zpools. Tip: set the arc_max value - by default zfs will use all physical ram for cache, set it to be sure you have enough ram left for any services you want running. Have you considered using one or both SSD drives with zfs? They can be added as cache or log devices to help performance. See man zpool under Intent Log and Cache Devices. This is a very interesting point. In terms if SSDs for cache, I was planning on using a pair of Samsung Pro 512GB SSDs for this purpose (which I haven't bought yet). But I tire of buying stuff, so I have a pair of 40GB Intel SSDs for use as sys disks and several Intel 160GB SSDs lying around that I can combine with the existing 256GB SSDs for a cache. Then use my 36x3TB for the beasty NAS. Agreed that 256G mirrored SSDs are kind of wasted as system drives. The 40G mirror sounds ideal. Update; I went with ZFS as I didn't want to confuse the toolset needed to support this server. Although gmirror is not hard to figure out, I wanted consistency in systems. So I've a booted 9.1 rel using a mirrored ZFS system disk. The drives do support TRIM but am unsure how this plays with ZFS. I did the standard partition scheme of; root@kronos:/root # gpart show = 34 78165293 da0 GPT (37G) 34 1281 freebsd-boot (64k) 162 6 - free - (3.0k) 168 83886082 freebsd-swap (4.0G) 8388776 697765443 freebsd-zfs (33G) 78165320 7 - free - (3.5k) = 34 78165293 da1 GPT (37G) 34 1281 freebsd-boot (64k) 162 6 - free - (3.0k) 168 83886082 freebsd-swap (4.0G) 8388776 697765443 freebsd-zfs (33G) 78165320 7 - free - (3.5k) At any rate, thank you for the replies, very much appreciate it. Especially since building a rather large production worthy NAS not knowing a lick of freeBSD. The reasons going with freeBSD are 2 fold; ZFS stability,seems a better marriage then ZOL. Correctly provides NFS pre attributes on write reply; mtime. Linux does not. While its a steep learning curve, the 2 points above require the use of freeBSD or alike. - aurf ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
You would in theory as from what i remember every zfs filesystem takes up 64 kb of ram, so the savings could be massive 8) On 16 July 2013 10:41, Shane Ambler free...@shaneware.biz wrote: On 16/07/2013 14:41, aurfalien wrote: On Jul 15, 2013, at 9:23 PM, Warren Block wrote: On Mon, 15 Jul 2013, aurfalien wrote: ... thats the question :) At any rate, I'm building a rather large 100+TB NAS using ZFS. However for my OS, should I also ZFS or simply gmirror as I've a dedicated pair of 256GB SSD drives for it. I didn't ask for SSD sys drives, this system just came with em. This is more of a best practices q. ZFS has data integrity checking, gmirror has low RAM overhead. gmirror is, at present, restricted to MBR partitioning due to metadata conflicts with GPT, so 2TB is the maximum size. Best practices... depends on your use. gmirror for the system leaves more RAM for ZFS. Perfect, thanks Warren. Just what I was looking for. I doubt that you would save any ram having the os on a non-zfs drive as you will already be using zfs chances are that non-zfs drives would only increase ram usage by adding a second cache. zfs uses it's own cache system and isn't going to share it's cache with other system managed drives. I'm not actually certain if the system cache still sits above zfs cache or not, I think I read it bypasses the traditional drive cache. For zfs cache you can set the max usage by adjusting vfs.zfs.arc_max that is a system wide setting and isn't going to increase if you have two zpools. Tip: set the arc_max value - by default zfs will use all physical ram for cache, set it to be sure you have enough ram left for any services you want running. Have you considered using one or both SSD drives with zfs? They can be added as cache or log devices to help performance. See man zpool under Intent Log and Cache Devices. __**_ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/**mailman/listinfo/freebsd-**questionshttp://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-** unsubscr...@freebsd.org freebsd-questions-unsubscr...@freebsd.org ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
not recommended anymore you should run SU+J if your version supports it On 17 July 2013 00:08, Nikos Vassiliadis nv...@gmx.com wrote: On 07/16/13 21:27, Johan Hendriks wrote: Op dinsdag 16 juli 2013 schreef Charles Swiger (cswi...@mac.com) het volgende: Hi-- On Jul 16, 2013, at 10:33 AM, Johan Hendriks joh.hendr...@gmail.com** javascript:; wrote: [ ... ] I would us a zfs for the os. I have a couple of servers that did not survive a power failure with gmirror. The problems i had was when the power failed one disk was in a rebuilding state and then when the background fsck started or was busy for some time it would crash the whole server. Well, don't do that. :-) When the server reboots because of a powerfailure at night, then it boots. Then it starts to rebuild the mirror on its own, and later the fsck kicks in. Not much i can do about it. You could add geom_journal which will minimize the time of fsck to a second or something like that. Then you don't have to use background fsck anymore. Actually geom_journal's manual page mentions an interesting side-effect of geom_journal over a geom_mirror: you can turn off component synchronization. Geom_journal will re-play last writes so whatever was changed just before the crash will be re-written to both disks. I haven't used this but it makes sense in theory. Maybe i should have done it without the automatic attachment for a new device. I always turn off automatic synchronization or stale components as well. It seems to me that people don't really use geom_journal or maybe they just don't talk about it like it's some sort of secret:) just my two cents, Nikos __**_ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/**mailman/listinfo/freebsd-**questionshttp://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-** unsubscr...@freebsd.org freebsd-questions-unsubscr...@freebsd.org ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On 16/07/2013 14:41, aurfalien wrote: On Jul 15, 2013, at 9:23 PM, Warren Block wrote: On Mon, 15 Jul 2013, aurfalien wrote: ... thats the question :) At any rate, I'm building a rather large 100+TB NAS using ZFS. However for my OS, should I also ZFS or simply gmirror as I've a dedicated pair of 256GB SSD drives for it. I didn't ask for SSD sys drives, this system just came with em. This is more of a best practices q. ZFS has data integrity checking, gmirror has low RAM overhead. gmirror is, at present, restricted to MBR partitioning due to metadata conflicts with GPT, so 2TB is the maximum size. Best practices... depends on your use. gmirror for the system leaves more RAM for ZFS. Perfect, thanks Warren. Just what I was looking for. I doubt that you would save any ram having the os on a non-zfs drive as you will already be using zfs chances are that non-zfs drives would only increase ram usage by adding a second cache. zfs uses it's own cache system and isn't going to share it's cache with other system managed drives. I'm not actually certain if the system cache still sits above zfs cache or not, I think I read it bypasses the traditional drive cache. For zfs cache you can set the max usage by adjusting vfs.zfs.arc_max that is a system wide setting and isn't going to increase if you have two zpools. Tip: set the arc_max value - by default zfs will use all physical ram for cache, set it to be sure you have enough ram left for any services you want running. Have you considered using one or both SSD drives with zfs? They can be added as cache or log devices to help performance. See man zpool under Intent Log and Cache Devices. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On 16/07/2013 10:41, Shane Ambler wrote: On 16/07/2013 14:41, aurfalien wrote: On Jul 15, 2013, at 9:23 PM, Warren Block wrote: On Mon, 15 Jul 2013, aurfalien wrote: ... thats the question :) At any rate, I'm building a rather large 100+TB NAS using ZFS. However for my OS, should I also ZFS or simply gmirror as I've a dedicated pair of 256GB SSD drives for it. I didn't ask for SSD sys drives, this system just came with em. This is more of a best practices q. ZFS has data integrity checking, gmirror has low RAM overhead. gmirror is, at present, restricted to MBR partitioning due to metadata conflicts with GPT, so 2TB is the maximum size. Best practices... depends on your use. gmirror for the system leaves more RAM for ZFS. Perfect, thanks Warren. Just what I was looking for. I doubt that you would save any ram having the os on a non-zfs drive as you will already be using zfs chances are that non-zfs drives would only increase ram usage by adding a second cache. zfs uses it's own cache system and isn't going to share it's cache with other system managed drives. I'm not actually certain if the system cache still sits above zfs cache or not, I think I read it bypasses the traditional drive cache. For zfs cache you can set the max usage by adjusting vfs.zfs.arc_max that is a system wide setting and isn't going to increase if you have two zpools. Tip: set the arc_max value - by default zfs will use all physical ram for cache, set it to be sure you have enough ram left for any services you want running. Have you considered using one or both SSD drives with zfs? They can be added as cache or log devices to help performance. See man zpool under Intent Log and Cache Devices. I agree with the sentiment of using the SSD as ZFS cache - it's possibly the only logical use for them. I guess that with 100Tb worth of Winchesters you're not on a very tight budget, and not too tight on RAM for the OS either. If I was going to do this I'd stick with the OS on UFS and a gmirror because I simply don't trust ZFS. This is based on pure prejudice and inexperience. I know how to arrange disks on a UNIX file system for performance - what to use for swap, where tmp files should go and so on. I also know where every file will be, physically, in the event of trouble. And here's the clincher: If the machine blows up I can simply take one of the mirrored drives, slap it in to some new hardware and I've got a very reasonable chance that it'll boot. Can I do this with ZFS? I get the feeling that the answer is an emphatic maybe. So all things considered, I'd need a good reason not to stick with what I know works reliably and can be recovered in the event of a disaster (UFS), but I'm happy to watch and learn from everyone else's experience! ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
Op dinsdag 16 juli 2013 schreef Frank Leonhardt (fra...@fjl.co.uk) het volgende: On 16/07/2013 10:41, Shane Ambler wrote: On 16/07/2013 14:41, aurfalien wrote: On Jul 15, 2013, at 9:23 PM, Warren Block wrote: On Mon, 15 Jul 2013, aurfalien wrote: ... thats the question :) At any rate, I'm building a rather large 100+TB NAS using ZFS. However for my OS, should I also ZFS or simply gmirror as I've a dedicated pair of 256GB SSD drives for it. I didn't ask for SSD sys drives, this system just came with em. This is more of a best practices q. ZFS has data integrity checking, gmirror has low RAM overhead. gmirror is, at present, restricted to MBR partitioning due to metadata conflicts with GPT, so 2TB is the maximum size. Best practices... depends on your use. gmirror for the system leaves more RAM for ZFS. Perfect, thanks Warren. Just what I was looking for. I doubt that you would save any ram having the os on a non-zfs drive as you will already be using zfs chances are that non-zfs drives would only increase ram usage by adding a second cache. zfs uses it's own cache system and isn't going to share it's cache with other system managed drives. I'm not actually certain if the system cache still sits above zfs cache or not, I think I read it bypasses the traditional drive cache. For zfs cache you can set the max usage by adjusting vfs.zfs.arc_max that is a system wide setting and isn't going to increase if you have two zpools. Tip: set the arc_max value - by default zfs will use all physical ram for cache, set it to be sure you have enough ram left for any services you want running. Have you considered using one or both SSD drives with zfs? They can be added as cache or log devices to help performance. See man zpool under Intent Log and Cache Devices. I agree with the sentiment of using the SSD as ZFS cache - it's possibly the only logical use for them. I guess that with 100Tb worth of Winchesters you're not on a very tight budget, and not too tight on RAM for the OS either. If I was going to do this I'd stick with the OS on UFS and a gmirror because I simply don't trust ZFS. This is based on pure prejudice and inexperience. I know how to arrange disks on a UNIX file system for performance - what to use for swap, where tmp files should go and so on. I also know where every file will be, physically, in the event of trouble. And here's the clincher: If the machine blows up I can simply take one of the mirrored drives, slap it in to some new hardware and I've got a very reasonable chance that it'll boot. Can I do this with ZFS? I get the feeling that the answer is an emphatic maybe. So all things considered, I'd need a good reason not to stick with what I know works reliably and can be recovered in the event of a disaster (UFS), but I'm happy to watch and learn from everyone else's experience! I would us a zfs for the os. I have a couple of servers that did not survive a power failure with gmirror. The problems i had was when the power failed one disk was in a rebuilding state and then when the background fsck started or was busy for some time it would crash the whole server. Removing the disk that was rebuilding resolved the issue. This happened to me more than once. Most of the times it worked as advertised but not always. Before people tell me to use an UPS, i used a UPS but the damn thing gave way itself. Then after it came back from the warranty repair it gave way again. Some times it came back right away, leaving some servers survive and some in the state they where. It was hard to find the cause in the beginning because of the fact some servers did survive the power failure. We did not suspect the UPS at first. Anyway, gmirror did not work for me in all cases. I am now running a few servers with a zfs root. I did not have any problems with them till now (knock on wood). Since reading that swap on zfs root can cause trouble i have a separate freebsd-swap partition for the swap. Gr Johan __**_ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/**mailman/listinfo/freebsd-**questionshttp://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On Jul 16, 2013, at 2:41 AM, Shane Ambler wrote: On 16/07/2013 14:41, aurfalien wrote: On Jul 15, 2013, at 9:23 PM, Warren Block wrote: On Mon, 15 Jul 2013, aurfalien wrote: ... thats the question :) At any rate, I'm building a rather large 100+TB NAS using ZFS. However for my OS, should I also ZFS or simply gmirror as I've a dedicated pair of 256GB SSD drives for it. I didn't ask for SSD sys drives, this system just came with em. This is more of a best practices q. ZFS has data integrity checking, gmirror has low RAM overhead. gmirror is, at present, restricted to MBR partitioning due to metadata conflicts with GPT, so 2TB is the maximum size. Best practices... depends on your use. gmirror for the system leaves more RAM for ZFS. Perfect, thanks Warren. Just what I was looking for. I doubt that you would save any ram having the os on a non-zfs drive as you will already be using zfs chances are that non-zfs drives would only increase ram usage by adding a second cache. zfs uses it's own cache system and isn't going to share it's cache with other system managed drives. I'm not actually certain if the system cache still sits above zfs cache or not, I think I read it bypasses the traditional drive cache. For zfs cache you can set the max usage by adjusting vfs.zfs.arc_max that is a system wide setting and isn't going to increase if you have two zpools. Tip: set the arc_max value - by default zfs will use all physical ram for cache, set it to be sure you have enough ram left for any services you want running. Have you considered using one or both SSD drives with zfs? They can be added as cache or log devices to help performance. See man zpool under Intent Log and Cache Devices. This is a very interesting point. In terms if SSDs for cache, I was planning on using a pair of Samsung Pro 512GB SSDs for this purpose (which I haven't bought yet). But I tire of buying stuff, so I have a pair of 40GB Intel SSDs for use as sys disks and several Intel 160GB SSDs lying around that I can combine with the existing 256GB SSDs for a cache. Then use my 36x3TB for the beasty NAS. - aurf ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
Op dinsdag 16 juli 2013 schreef Charles Swiger (cswi...@mac.com) het volgende: Hi-- On Jul 16, 2013, at 10:33 AM, Johan Hendriks joh.hendr...@gmail.comjavascript:; wrote: [ ... ] I would us a zfs for the os. I have a couple of servers that did not survive a power failure with gmirror. The problems i had was when the power failed one disk was in a rebuilding state and then when the background fsck started or was busy for some time it would crash the whole server. Well, don't do that. :-) When the server reboots because of a powerfailure at night, then it boots. Then it starts to rebuild the mirror on its own, and later the fsck kicks in. Not much i can do about it. Maybe i should have done it without the automatic attachment for a new device. Seriously, bring up the box on one disk, force a foreground fsck if needed to get the filesystem to known clean state, and then rebuild the mirror. Mixing the mirror rebuild with something like an fsck will just thrash the disks. [ ... ] Before people tell me to use an UPS, i used a UPS but the damn thing gave way itself. Then after it came back from the warranty repair it gave way again. Grr. That's when you want find another UPS vendor. Is apc not the right choice? I think i got a monday morning model. Some times things fail! Regards, -- -Chuck ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On Tue, 16 Jul 2013, aurfalien wrote: On Jul 16, 2013, at 2:41 AM, Shane Ambler wrote: I doubt that you would save any ram having the os on a non-zfs drive as you will already be using zfs chances are that non-zfs drives would only increase ram usage by adding a second cache. zfs uses it's own cache system and isn't going to share it's cache with other system managed drives. I'm not actually certain if the system cache still sits above zfs cache or not, I think I read it bypasses the traditional drive cache. For zfs cache you can set the max usage by adjusting vfs.zfs.arc_max that is a system wide setting and isn't going to increase if you have two zpools. Tip: set the arc_max value - by default zfs will use all physical ram for cache, set it to be sure you have enough ram left for any services you want running. Have you considered using one or both SSD drives with zfs? They can be added as cache or log devices to help performance. See man zpool under Intent Log and Cache Devices. This is a very interesting point. In terms if SSDs for cache, I was planning on using a pair of Samsung Pro 512GB SSDs for this purpose (which I haven't bought yet). But I tire of buying stuff, so I have a pair of 40GB Intel SSDs for use as sys disks and several Intel 160GB SSDs lying around that I can combine with the existing 256GB SSDs for a cache. Then use my 36x3TB for the beasty NAS. Agreed that 256G mirrored SSDs are kind of wasted as system drives. The 40G mirror sounds ideal. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
Hi-- On Jul 16, 2013, at 10:33 AM, Johan Hendriks joh.hendr...@gmail.com wrote: [ ... ] I would us a zfs for the os. I have a couple of servers that did not survive a power failure with gmirror. The problems i had was when the power failed one disk was in a rebuilding state and then when the background fsck started or was busy for some time it would crash the whole server. Well, don't do that. :-) Seriously, bring up the box on one disk, force a foreground fsck if needed to get the filesystem to known clean state, and then rebuild the mirror. Mixing the mirror rebuild with something like an fsck will just thrash the disks. [ ... ] Before people tell me to use an UPS, i used a UPS but the damn thing gave way itself. Then after it came back from the warranty repair it gave way again. Grr. That's when you want find another UPS vendor. Regards, -- -Chuck ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
Hi-- On Jul 16, 2013, at 11:27 AM, Johan Hendriks joh.hendr...@gmail.com wrote: Well, don't do that. :-) When the server reboots because of a powerfailure at night, then it boots. Then it starts to rebuild the mirror on its own, and later the fsck kicks in. Not much i can do about it. Maybe i should have done it without the automatic attachment for a new device. It's normally the case that getting a hot spare automatically attached should be fine, but not if you also have the box go down entirely and need to fsck. I'm more used to needing to explicitly physically swap out a failed mirror component, in which case one can make sure the system is OK before the replacement drive goes in. [ ... ] Before people tell me to use an UPS, i used a UPS but the damn thing gave way itself. Then after it came back from the warranty repair it gave way again. Grr. That's when you want find another UPS vendor. Is apc not the right choice? I think i got a monday morning model. Some times things fail! APC is decent for desktops, but I'm dubious about them when it comes to entire racks or a DC. I like Leviton's PDUs/MDUs and TVSS; for a medium-sized UPS (10-40 kVA) Liebert and PowerWare (now Eaton) were good. Liebert's PDUs are also pretty good. Regards, -- -Chuck PS: I ran a small DC in NYC with a 20kVA PowerWare 9330 behind a Leviton 57000 TVSS; the Cupertino locals have ~650kVA worth of Bloom boxes and a Cummins diesel genset as a backup just for this building. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On 07/16/13 21:27, Johan Hendriks wrote: Op dinsdag 16 juli 2013 schreef Charles Swiger (cswi...@mac.com) het volgende: Hi-- On Jul 16, 2013, at 10:33 AM, Johan Hendriks joh.hendr...@gmail.comjavascript:; wrote: [ ... ] I would us a zfs for the os. I have a couple of servers that did not survive a power failure with gmirror. The problems i had was when the power failed one disk was in a rebuilding state and then when the background fsck started or was busy for some time it would crash the whole server. Well, don't do that. :-) When the server reboots because of a powerfailure at night, then it boots. Then it starts to rebuild the mirror on its own, and later the fsck kicks in. Not much i can do about it. You could add geom_journal which will minimize the time of fsck to a second or something like that. Then you don't have to use background fsck anymore. Actually geom_journal's manual page mentions an interesting side-effect of geom_journal over a geom_mirror: you can turn off component synchronization. Geom_journal will re-play last writes so whatever was changed just before the crash will be re-written to both disks. I haven't used this but it makes sense in theory. Maybe i should have done it without the automatic attachment for a new device. I always turn off automatic synchronization or stale components as well. It seems to me that people don't really use geom_journal or maybe they just don't talk about it like it's some sort of secret:) just my two cents, Nikos ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
to gmirror or to ZFS
... thats the question :) At any rate, I'm building a rather large 100+TB NAS using ZFS. However for my OS, should I also ZFS or simply gmirror as I've a dedicated pair of 256GB SSD drives for it. I didn't ask for SSD sys drives, this system just came with em. This is more of a best practices q. Thanks in advance, - aurf ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On Mon, 15 Jul 2013, aurfalien wrote: ... thats the question :) At any rate, I'm building a rather large 100+TB NAS using ZFS. However for my OS, should I also ZFS or simply gmirror as I've a dedicated pair of 256GB SSD drives for it. I didn't ask for SSD sys drives, this system just came with em. This is more of a best practices q. ZFS has data integrity checking, gmirror has low RAM overhead. gmirror is, at present, restricted to MBR partitioning due to metadata conflicts with GPT, so 2TB is the maximum size. Best practices... depends on your use. gmirror for the system leaves more RAM for ZFS. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: to gmirror or to ZFS
On Jul 15, 2013, at 9:23 PM, Warren Block wrote: On Mon, 15 Jul 2013, aurfalien wrote: ... thats the question :) At any rate, I'm building a rather large 100+TB NAS using ZFS. However for my OS, should I also ZFS or simply gmirror as I've a dedicated pair of 256GB SSD drives for it. I didn't ask for SSD sys drives, this system just came with em. This is more of a best practices q. ZFS has data integrity checking, gmirror has low RAM overhead. gmirror is, at present, restricted to MBR partitioning due to metadata conflicts with GPT, so 2TB is the maximum size. Best practices... depends on your use. gmirror for the system leaves more RAM for ZFS. Perfect, thanks Warren. Just what I was looking for. - aurf ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org