Slashdot Mirror
Linux LVM - Is It Ready for Prime Time?
Deagol asks: "I'd like to replace our aging IBM server with a commodity solution (Linux, 3Ware cards, and lots of IDE drives). The main reason is price (the cost of 5 36GB SCSI disks for this sucker -- one of which died today -- could pay for the replacement server with 2TB of usable space after RAID-5. Being a huge fan of AIX's LVM,I've recently been playing with the Linux version of LVM. It's got all the right features (and even the ability to shrink logical volumes, a feature which AIX 4.3.3 doesn't have!), though the commands aren't as polished as the AIX counterparts. The big question for me is, will it stand up and be stable under heavy load, like the IBM does? Is anyone running Linux LVM on a 1TB+, 24/7 production machine?"
For those who haven't used it, the LVM support on AIX is probably the best in the industry. It allows you to do just about anything, in a very clean, structured (if somewhat hard to figure out at first) way. I personally have not used it on Linux, so can't comment on the poster's main question, but the kernel-raid stuff is not up to part with IBM's LBM.
"He who would learn astronomy, and other recondite arts, let him go elsewhere. " -- John Calvin, commenting on Genesis 1
-The main reason is price (the cost of 5 36GB SCSI disks for this sucker -- one of which died today --
Odds are the drives are OEM versions of a very popular drive vendor, perhaps pop it out, figure what kind of drive it is, buy a new one that is an exact match (or better yet, buy five new ones of exactly the same type) and replace them yourself for +/- $2,000 total) and restore from your backup. Maybe this is a little oversimplified but if it is a RS/6000 box odds are it uses regular ol' SCSI drives.
Glonoinha the MebiByte Slayer
I think that was EVMS, not LVM. EVMS was a potential IBM replacement for LVM, but was rejected by Linus.
include $sig;
1;
I built LVM into a RH 7.3 kernel and used it for a DB2 database box. Worked great with the dell 2650 and a 1 tb powervault 220s (raid5 ...aaarrrg...) with the perc3 raid controller. It passed our rather aggressive load testing cyle, ~7 days constant load w/ a variety of different tests, (broken queries, massive table joins, etc, stuff you WOULDN'T want to see in production) and passed with flying colors. It remained in production for a little under a year until we migrated to a Oracle on Linux solution w/o LVM (RedHat AS2.1). I'd do it again in a minute if the implementation called for it.....
just my 2 cents.......
But, it's looking as if the LVM code isn't actually included in the 2.5/2.6 series of kernel (I could be wrong). If you plan on upgrading to this eventually, stay away from LVM. If you don't care just dive in.
Well, according to SCO, the LVM support in Linux was added by IBM, so it's probably pretty good.
(/me ducks)
"Lawyers are for sucks."
- Doug McKenzie
So the poster will be able to replace his costly IBM hardware with free IBM software. If that's not ironic, I don't know what is.
Nothing to see here; Move along.
I have been using LVM on my workstation and just about every machine that I have built in the past 2 years. I also use AIX everyday and agree that the LVM is very robust but I prefer the simplicity of the linux LVM.I recently built a box with software raid 0 and made the whole thing on physical volume and have had zero problems. I have also user the LVM to migrate date from one dist to another without any problems. Oh and did I mention that the LVM howto is really really easy to understand even if you have no previous LVM experiance.
is that they keep replacing it and reimplementing it in the kernel.
The one linked to in the article (Sistina's) is in 2.4. I'm using it at home, and I like it. We're considering using it at work, but I hear rumours that 2.6 will contain Something Completely Different (Again), which annoys me.
You cannot apply a technological solution to a sociological problem. (Edwards' Law)
As far as stability and reliability goes, I haven't experienced any problems yet.
/boot has to be a non-LVM partition. I'd recommend making a root filesystem of a few hundred megs, outside of LVM. It's less of a hassle than making a separate /boot filesystem outside of LVM.
... etc.) but if you're used to AIX LVM, you'll be up to speed with this in no time.
There are some features though that are still missing from Linux LVM, compared to AIX LVM. One of them is mirroring on the logical volume level (no mklvcopy command). You can sort of get around this by creating a software raid device, and then make it a physical disk. Or even better: just go for a hardware RAID(1/5) solution.
Another thing to keep in mind is that, unlike in AIX, you can't put *all* filesystems in LVM. Either the root filesystem or
Also if you want to be able to resize live filesystems, you have to be careful about your choice of filesystem. Reiserfs for example supports online resizing, while ext3 doesn't (yet?)
All things considered Linux LVM is a great addition to Linux, but it's not as nicely integrated yet as AIX's LVM.
One final thing to note is that the Linux LVM commands seem to be modeled after HP-UX LVM rather than AIX LVM. (e.g. lvcreate instead of mklv, vgdisplay instead of lsvg
The mail spool for a 15k-user ISP in southwestern Ontario is running on Slack9 + LVM (Reiserfs). It exports the spool via NFS and the edge servers (SMTP+IMAP4+POP3, virus+spamscan) mount the spool directly over ipsec. No issues. I can grow the filesystem, take snapshots and it all just works. The PostgreSQL database is also on an LVM volume, but I haven't had to do much with it related to LVM yet, as pg_dump works live.
I have a number of other mail spools for businesses around the area (probably a half dozen to 10 or so) -- around 100-300 users each. Same story, backups are very nice when you can just take a snapshot.
I'm currently evaluating Appgen Custom Suite as a replacement for our current Accpac and Misys accounting and inventory/manufacturing infrastructure. LVM is very nice here for growing the db partition and snapshotting means the system is only down for seconds instead of the time it'd take to dump the entire db to tape.
All in all, I am very pleased with LVM. I do all my LVM on software RAID1 or hardware RAID5, but there's been no interaction or badness show up so far. I realize it's possible to do RAID1 with LVM but there's no documentation on how to "un-fail" the volume after a drive fails or how to manage it, so until then md or the hardware tools will have to do. :-)
The underlying hardware of the replacement box is 16 180GB IDE drives, split between 2 8-port 3Ware cards. Each card is doing RAID-5 with 7 drives, the 8th being a hot spare. I let the hardware manage the redundancy for me.
Linux actually sees 2 1080GB (~1TB) SCSI drives. Using standard methods, I can partition these 2 drives any way you normally can under linux. But that kind of binds my feet.
For example, what if the 500GB filesystem for Group A is full and they're clamoring for more space? Using standard partitioning, I'd have to create another larger filesystem, copy the data over, re-export the space, then finally re-claim the old filesystem. In addition to being a pain in the ass, this would require a down time for the users. However, using LVM, I can simply append more of those "chunks" I mentioned above, without creating a new filesystem, Better yet, I can do this on a live system -- my users won't notice a thing (other than the sudden appearance of new file space).
It's really cool stuff.
I'm still uneasy about the reliability, though. I have no hesitation about resizing on-the-fly with hundreds of people running batch jobs on the filesystem under AIX. I haven't seen enough of the Linux LVM in action yet to be that confident of its abilities. If anyone can comment on how Linux NFS and Samba handle having the underlying filesystem resized on-the-fly, I'd love to hear about it.
Method of processing duck feet
I'm not going to buy into the IBM LVM vs Linux Software Raid debate, but no one has mentioned something thatâ(TM)s just as important. One of the big advantaged with a good SCSI enclosure is the ability to pull and pop drives in and out without powering down.
With good hardware, you can walk up to a running machine and replace the failed drive then and there. Hopefully your 144Gb raid-5 array has been fully rebuilt by the time you come back from lunch. If you don't have hot-swap hardware, you have to schedule downtime, come back later that night, shut it down, pull the drive and pop in a new one. And hope everything powers up OK, cos if the power supply stuffs up at that time of night and you don't have a (good) support contract you are going to have a lot of fun getting everything going again before the rest of the office shows up for work.
I know you can get hot-swap IDE hardware these days, but I've never used them. I suspect hot-swap IDE drives are not that much cheaper then SCSI, but I could be wrong.
One last little bit of advice, try including a hot spare in your array. Its nice to come in in the morning and read an email saying that a hard drive failed last night, and the array was automatically re-built using the spare before start of business. If you are going to go with non hot-swap hardware, Iâ(TM)d say this is a must. Running raid-5 in degraded mode is no fun.
Democracy isn't about no one telling you what to do. It's about everyone telling you what to do.
Been around pretty long.
lvextend to increase the volume.
resize_reiserfs to increase the filesystem.
Only works if you want to increase the filesystem, not shrink it.
Je ne parle pas francais.