Forum OpenACS Q&A: Switching to Raid 1...

Collapse
Posted by David Kuczek on
Right now I am running oacs 3.2.5 on the following system without a raid at all:

RH8.0
Dual Processor PIII 1Ghz
1GB Ram
20GB IBM Hard Drive (IDE)

Since I recently had a pretty nasty hard drive crash, I am planning to switch to a Raid 1 system with 2 120GB IDE hard drives...

What raid controller should I use? Hardware, software? If hardware, which product goes well and reliable with RH8.0?

Which hard drives are most reliable? I will most probably not be able to afford scsi...

Thanks

Collapse
2: Re: Switching to Raid 1... (response to 1)
Posted by Ola Hansson on
I have only good things to say about software raid and would recommend it. I wrote a little guide to doing software raid1 on Debian that may or may not be useful to you on RH 8.0, but let me link to it anyway for future readers of this thread...

http://infogettable.net/linux/reiserfs-raid1/

/Ola

Collapse
3: Re: Switching to Raid 1... (response to 1)
Posted by Bruno Mattarollo on
Hi David,

We have been using RAID 1 (software) with RedHat 7.3 and 8.0 without any problems ... That's the cheapest solution you can get :)

We are using it on a server that is running a quite active PostgreSQL database and it's running very nicely.

I couldn't provide you with performance details to compare software vs. hardware RAID though but I haven't seen a particular slowdown compared to the server as it was before we did implement the RAID.

Collapse
4: Re: Switching to Raid 1... (response to 1)
Posted by Don Baccus on
I use software RAID 1, too, if that's any help to you, on a dual PIII 450, and it requires very little in the way of CPU resources.  And you've got two processors that are just a bit faster than mine ...
Collapse
5: Re: Switching to Raid 1... (response to 1)
Posted by Malte Sussdorff on
Okay, here are our experiences: Stay away from the cheap hardware controllers like DawiControl (both chipssets they use). Have a look at the CT (german computing magazine) artikel on that (don't remeber the issue, was fairly recent). They just don't work that well with Linux. Now we use 3ware and run it in RAID 5 configuration. Works really nice, no problems with kernels and good software comes with it. But at 350 EUR it is not cheap for a four channel IDE-RAID either.
Collapse
7: Re: Switching to Raid 1... (response to 1)
Posted by Jun Yamog on
Malte is right.  Stay away from cheap IDE RAID.  They are actually more of just a custom IDE controller and a device driver that has software RAID.  If you can't afford the expensive cards then just use Linux software raid.

http://www4.tomshardware.com/storage/20021112/index.html

Collapse
6: Re: Switching to Raid 1... (response to 1)
Posted by C. R. Oldham on
David,

Here are some things to think about when considering RAID 1 options.

  1. Most onboard IDE/ATA controllers were not designed for fault tolerance. If you put both drives on the same controller (master/slave) failure of one drive will most likely render the other drive inaccessible to the controller. Thus you really need one controller per disk. In addition, even in a one controller per disk situation a drive failure could also make the controller fail and bring the whole box down.
  2. Until recently you needed some mojo to make your root partition RAID with software RAID. I think late versions of lilo/grub/linux kernel support root on RAID.
  3. If you decide to go the hardware route, Promise does make some cheap ATA RAID controllers that do "almost" hardware RAID. A lot of the work is done in the driver software, but still the whole array is presented to the kernel as one device. I have a couple of these and they work OK. What I've never investigated is if the driver presents an interface in /proc that would let you monitor the status of the disks. The Tech Report did a really in-depth review and analysis of different hardware IDE RAID controllers--it is here.
  4. "Good" RAID 1 implementation should improve read performance since technically data can be read from two drives at once (unless the array is missing a disk). I remember rumblings on the linux-raid list that the RAID implementation does do this; I don't know about hardware solutions.
  5. Hot swapping of IDE disks is not available unless your controller supports it and you have the right IDE cages. The IDE electrical spec was never designed for hot swap, so be careful when choosing hot-swap cages.
  6. 3ware cards are the acknowledged leader in IDE RAID. They can be expensive, but offer all of the features of a SCSI solution (hot swap/hot spare). Good Linux support too.

I have been looking at IDE RAID options for some time. Since IDE disks do not come in more than 7200 RPM version yet but I can get SCSI at 15000 RPM an IDE solution will probably not reach the performance of a SCSI option. You're looking at probably half the cost for IDE though, and more total storage with less hardware since you can get 250 GB IDE disks. In the end we will still go with SCSI I think.

Collapse
8: Re: Switching to Raid 1... (response to 1)
Posted by David Kuczek on
Hey Bruno,

can you direct me to a howto on installing software raid 1 on RH 8.0? What kind of hard drives do you guys use? ide or scsi?

Collapse
9: Re: Switching to Raid 1... (response to 1)
Posted by Don Baccus on
I'm SCSI myself.  When you install Linux these days, and set up your partitions, all you need to do is to tell it to turn two partions into a RAID 1 mirror and off you go, everything's set up automatically, it's no harder than setting up a normal partition.

I've not added additional disks to the system so don't know what's involved in this case.

Collapse
Posted by Bruno Mattarollo on

Hello David,

Don's right, it's very simple indeed ... I have seen our Master-Guru-Linux-I've-Never-Used-Windows-consultant do it and it was a 2.5 minutes work (if you know what partitioning you want). We are using IDE on the postgreSQL server and SCSI on an LDAP server.

With the speed of IDE drives and ATA controllers and the difference in cost from IDE to SCSI I would use IDE unless you are planning on a massive server that will need high throughoutput... At least IDE is working for us :)

btw this link might help:

http://www.redhat.com/docs/manuals/linux/RHL-8.0-Manual/admin-primer/s1-storage-raid.html.

Collapse
Posted by Rich Graves on
We have had far more downtime and lost data due to failed hardware RAID controllers
than to failed disk drives.

Unless you're talkign about a pre-integrated system like a NetApp or can prove that your system is seriously CPU-bound,
use software RAID. The fact that you can plug half your
mirror into any system and have it work is priceless.

OK, all of my experience is with SCSI, I know IDE takes
more CPU, so maybe it makes some sense there. But I don't
think the supposed benefits of hardware RAID outweigh the risk you'll lose everything if the totally proprietary layout of your disks goes haywire.

Yeah, with software RAID you usually lose the ability to
hot-swap a replacement disk. In practive I found that feature useful
and functional ONCE. I've had disks in large software RAID
systems fail 3 or 4 times and have never had problems
scheduling a quick reboot to swap drives. Usually it's time
for a kernel upgrade anyway. The actual RAID rebuild can be
done while the system is running.

Collapse
Posted by Andrew Piskorski on
So far I am quite happy with the 3Ware Escalade 6410 (32 bit) 4 drive IDE RAID card I'm using on my home system, doing RAID 10. However, now that you can only buy the 64 bit cards, the prices are about twice as much. Probably around $200+ for a 4 drive card instead of the $100+ I paid for the older 32 bit card. Set up is very, very easy. I haven't tried using software RAID yet on any of my Debian systems so I can't compare how easy/difficult that is.

Supposedly, performance wise, Linux software RAID is just as fast or faster than most hardware cards for RAID 0 and 1, but I'm not sure about RAID 5 and 10. I also haven't done any serious benchmarking to see what the RAID 10 really buys me for running an RDBMS or the like, but it should be much faster than RAID 1.

Collapse
Posted by Andrew Grumet on
David K wrote
can you direct me to a howto on installing software raid 1 on RH 8.0? What kind of hard drives do you guys use? ide or scsi?
Take a look at the RedHat8 customization guide section on software RAID config, http://www.redhat.com/docs/manuals/linux/RHL-8.0-Manual/custom-guide/ch-software-raid.html.

I recently set up an RH8 system with raid1 on two 120GB IDE disks. I did this at install time using Disk Druid. The entire disk, including boot partition, is raided. It was rat simple.

One gotcha is that, by default, grub is only installed on the MBR of one of the two disks. So if that disk dies, you won't be able to boot from the other one. If this worries you, it is possible to install grub to the MBR of the other disk as well, using the instructions on the grub website.

Collapse
Posted by David Kuczek on
Andrew,

silly question, but just to make sure: do you install grub to the second MBR after RH8.0 has finished its installation process? Did you install it on your second drive? Probably yes to both...

What kind of hard drives did you use and why?

Collapse
Posted by Lars Pind on
Have any of you people running software RAID ever experienced a hard drive crash which you were able to recover thanks to the RAID configuration? I'm just curious, since that would be my object with running RAID 1 in the first place.

I just inquired about software RAID over at rackspace, with whom we host our server, and they said that they don't offer it:

"Currently we do not configure software raids. [...] we honestly do not recommend software raid as it is routinely unstable and doesnt provide the same redundancy as hardware raid."

"I would like to reiterate the point that the reason we do not offer software RAID solutions is because they are unstable and not as effective as hardware RAID."

Are they just plain wrong about this?

/Lars

Collapse
Posted by Jun Yamog on
Interesting, using IDE HD on SCSI controller

http://www.tomshardware.com/storage/20030130/index.html

Collapse
Posted by Mike Sisk on
I'd consider a hardware RAID solution if at all possible.

Our primary backup machine is a dual-Athlon machine with a 3Ware IDE RAID card with 8 160-GB Maxtor disks (6 in RAID 5 and two hot spares  in 3Ware hot swap cages). Works great.

Our primary servers are all Dell rackmounts with Adaptec hardware SCSI RAID; some are straight mirrors but most are RAID 5--mostly using 15k rpm Seagate drives (stay away from recent vintage IBM SCSI drives...I had three fail last year).

Some of our Oracle databases are quite large and the backup machine has no problem keeping up with the dumps from the 15k rpm SCSI based machines--the data trasfers are limited by the 100-Mbit/sec network (gotta look into gigabit Ethernet one of these days...).

I'm very impressed with the performance of the 3Ware cards and they're well supported in Linux and Freebsd.

I had an older machine at one point using software RAID 5 on 4 IDE disks on a single controller--not optimal but it ran fine for years. Performance degraded during substained data transfers (this was a stats processing machine) but then it only had a single 366-MHz Celeron CPU.

BTW, the trend in Enterprise computing is software RAID. The setiathome folks awhile back reported they switched to software RAID on their Solaris machines after a hardware RAID failure resulted in a total data loss situation. The hardware architecture of these sorts of machines are much better suited to software RAID than PC-architecture with a single IDE controller.

Collapse
Posted by Ola Hansson on
Lars,

I have never experienced a hard drive crash using a RAID array, but I have _simulated one_ by powering off a box running with a working mirror, unplugging the power cord to one of the (IDE) drives and taking the system back up again. The mirror will "heal" transparently in the background; all I have to do is say:

raidhotadd /dev/md0 /dev/hda1
raidhotadd /dev/md1 /dev/hda2
raidhotadd /dev/md2 /dev/hda3
In reality, though, what happens in my case is that if I unplug hda (master no. 1, first half of mirror) and pretend it is broken, the BIOS of my mobo is too dumb to realize it should boot off of hde (master no. 2, second half of the mirror)... My guess is there are BIOSs that can handle this issue, though.

As long as the working part of the mirror is placed in the first IDE slot and a new replacement drive (of aproximately the same size) is in place (it too as a master), the box will boot, assuming lilo (or grub) has been correctly installed in the boot sector or boot partition of BOTH of the original drives. (The new drive will be bootable as the partitions get copied to it during healing)

Software RAID, too (just like most hardware RAID-solutions I,ve read about), lets you use n number of "hot spares" that will jump in transparantly in case of a failure in the mirror...

Collapse
Posted by Ola Hansson on
Rereading my post it sounds like the box won't come up unless there are two drives in place - which is not needed, of course.

Having just one half of the mirror in good shape will do to boot...

Collapse
Posted by Frank N. on

Lars, I think you might have to differentiate between two different levels of reliability here, not to mention SCSI vs. IDE.

The first is 'stay up at all costs'. In my experience all bets are off if one or more of your IDE disks connected to the mainboard are in questionable condition. Wether the system will stay up or reboot reliably from a BIOS/hardware point of view if one of the master IDE disks decides to call it a day is anyone's guess.

Thus for a hosting compagny with hundreds of disks, the use of SCSI hotswap RAID controllers will help ensure undisturbed sleep, because the, possibly partial, failure of one disk in a mirror array has predictable results: None in the near future. The admins will be notified by the monitoring software, and can replace the failed disk when convenient.

The second part is 'filesystem integrity'. I am looking forward to playing with FreeBSD's very capable Vinum Volume Manager. If my experience with the resilience of the UFS *BSD filesystem with enabled softupdates in the face of power failures and end user ignorance is any indication, then I will be very surprised if this 'software RAID' system is anything but very reliable from a data integrity point of view. I do not have much experience with Linux software RAID though.

So by using software IDE RAID/Vinum I would expect not to loose much data up to the time of the disk failure, but I would expect to have to manually unplug the dead/dying disk to bring the system up again. Wether that is acceptable for a given application will have to be balanced against the cost of a hardware RAID SCSI array.

PS: I will not be home much for the next five days or so.

Collapse
Posted by Frank N. on
David, I forgot a datapoint for one of your original questions.

I admin around 50 Seagate Barracuda IVs of the 40GB variety, around 15 of those are running 24/7, the rest are just desktops. I have never had one die on me, nor did I ever receive a DOA specimen, something that I cannot say about IBM DeathStars. 😟

Do note however that I have heard that Barracuda IVs loose around 75% of their speed if connected to a hardware RAID controller. Don't know wether this is true.

Something else: unless you really, really need the 120GB, then by all means use the smallest size disk you can get away with. The larger the disk capacity, the higher the power consumption. Heat ~ temperature ~ 1/reliability. This is especially true if you do not actively cool your disks with dedicated fans, can space them adequately etc. For the Barracuda IVs there is a definite step in power consumption in going from 40 to 60 GB, as the latter is a two platter design requiring a stronger motor.

Collapse
Posted by Andrew Grumet on
David,

Yes to both.

We've got two identical Western Digital drives of the 1200BB variety.  We didn't buy them for any special reason, other that that they were readily available at PCs For Everyone, a local retailer.