Best hard disks for Microserver

[weeboot]

http://www.ebuyer.com/store/Servers/cat/Server-Mem...

[HantsRat]

weeboot said:

http://www.ebuyer.com/store/Servers/cat/Server-Mem...

Any non HPE RAM at reasonable prices?

[buggalugs]

Murph7355 said:

alock said:

A large RAID 5 still makes sense for redundancy, i.e. you need to keep the service running during the day and then maybe restore from a backup out of hours.

RAID 5 does not make sense for protecting data in any way. Statistically a large rebuild will fail. You are lucky if it doesn't.
http://www.smbitjournal.com/2012/05/when-no-redund...

If a disk fails in a RAID 5, you are far better off running an incremental backup to update your last full backup. This will apply the least load on the remaining disks. If you cannot do this then a full backup is still a better option than a re-build.

Personally I would also avoid a RAID 1 on a home NAS as well. A hot-backup disk (not directly exposed to the network) with a daily or weekly synchronization covers you for the more common home scenarios such as accidental deletions and encrypting malware.

Hmmmm. It's getting late and I've had a few glasses of vino but something's not right in the logic used in that article.

First of all, by the logic used I would definitely have had a terminal failure at home, let alone witnessed shed load of others in a professional context over the last 25yrs. I haven't. I've had cause to rebuild my home arrays 4, possibly 5 times in the time I've been using NAS units at home and never once had issues. With the stats he's noting, with that sort of luck I should do the lottery tomorrow

Equally I don't recall ever having a catastrophic "URE" either. In day to day use or during a rebuild.

My home NAS units only have 4 drives apiece, and critical data is mirrored between them. I don't know of any places using dozens of disks in a single RAID 5 array, and would agree that starts to make not much sense!

I'll have a better read later.

I agree 100% that a RAID array is no replacement for a backup regime. But I'm not really sold on the thrust of the arguments in the article

I agree, everyone quotes that same article these days when discussing RAID 5 but it just doesn't marry with reality for me at all. I've never had a failiure rebuilding RAID5 yet according to that article you're lucky if it ever works. The URE issue would also apply to RAID1 and single drives. Most controllers do periodic consistency checks these days which again should constantly fail, but they don't. Article is wrong as far as I'm concerned.

Edited by buggalugs on Tuesday 13th June 16:07

[budgie smuggler]

buggalugs said:

I agree, everyone quotes that same article these days when discussing RAID 5 but it just doesn't marry with reality for me at all. I've never had a failiure rebuilding RAID5 yet according to that article you're lucky if it ever works. The URE issue would also apply to RAID1 and single drives. Most controllers do periodic consistency checks these days which again should constantly fail, but they don't. Article is wrong as far as I'm concerned.

Edited by buggalugs on Tuesday 13th June 16:07

I think the risk is over stated by that article but it is a risk, and the more spindles in the array, the bigger that risk is.

[buggalugs]

budgie smuggler said:

buggalugs said:

I agree, everyone quotes that same article these days when discussing RAID 5 but it just doesn't marry with reality for me at all. I've never had a failiure rebuilding RAID5 yet according to that article you're lucky if it ever works. The URE issue would also apply to RAID1 and single drives. Most controllers do periodic consistency checks these days which again should constantly fail, but they don't. Article is wrong as far as I'm concerned.

Edited by buggalugs on Tuesday 13th June 16:07

I think the risk is over stated by that article but it is a risk, and the more spindles in the array, the bigger that risk is.

Getting out of bed in the morning is a risk but I'm not going to change my behaviour because of it, it's the magnitude of the risk that's important. This article has some fundamentaly wrong assumptions in it, otherwise the internet would be full of people saying their arrays had failed, not full of people quoting articles.

[TonyRPH]

alock said:

http://www.smbitjournal.com/2012/05/when-no-redund...

article said:

We are talking about a thirty six terabyte array – which sounds large but this is a size that someone could easily have at home today, let alone in a business.

How many of you have a 36TB array at home?

[Murph7355]

buggalugs said:

Getting out of bed in the morning is a risk but I'm not going to change my behaviour because of it, it's the magnitude of the risk that's important. This article has some fundamentaly wrong assumptions in it, otherwise the internet would be full of people saying their arrays had failed, not full of people quoting articles.

This.

[deckster]

Murph7355 said:

buggalugs said:

Getting out of bed in the morning is a risk but I'm not going to change my behaviour because of it, it's the magnitude of the risk that's important. This article has some fundamentaly wrong assumptions in it, otherwise the internet would be full of people saying their arrays had failed, not full of people quoting articles.

This.

Maybe, maybe not. I only have a sample size of one, but I am running at 100% failure of rebuilding RAID5 arrays, and that was only a 6TB array.

Perhaps it's overstated but that doesn't mean that it's (a) not real, and (b) easily mitigated by using eg RAID 10. True you lose some storage space but hey, disks are cheap.

[budgie smuggler]

buggalugs said:

Getting out of bed in the morning is a risk but I'm not going to change my behaviour because of it, it's the magnitude of the risk that's important. This article has some fundamentaly wrong assumptions in it, otherwise the internet would be full of people saying their arrays had failed, not full of people quoting articles.

There speaks somebody who has never had a rebuild fail

There is nothing at all fundamentally wrong in the article, it's just that we are lucky enough that drives tend to outperform their spec'd URE rate.

[buggalugs]

Oh I've luck'ed myself out of one or two RAID situations in the past

I think a big thing is to make sure periodic scans are enabled otherwise you can develop bad sectors across the drives and not know about it until you come to read them, i.e. rebuild. If you're scanning monthly then it should be the scans that find failing sectors as they happen.

Also, what does URE mean? I haven't seen it defined anywhere. If you read the same sector again and again does that metric apply? Or is it the chance of any sector going bad in some period of time? Is it actually the write that doesn't work properly leading to the read failiure? Or is it a read head signal thing that means that it's just got to wait a few ms for the disc to spin round and try the sector again? The stat's meaning it just taken at an assumed face value. Not to mention that it hasn't changed much since sata was invented which makes me suspect that its' a figure someone plucked out of the air once.

Also why doesn't this apply to RAID 1 - if you're supposed to average an 'unrecoverable error' for every 12TB read (really??) why wouldn't a 12TB RAID10 rebuild fail just as often? Within an order or magnitude anyway.

Anyway that's just my thoughts on it, I'm not going to go all nerdrage on it it just seems like a storm in a teacup to me.

[sjg]

I've lived through enough RAID failures at employers and clients to not want that at home. That was proper enterprise controllers and disks too - not huge SATA disks and weak consumer controllers. In the space I work in there's been a big shift away from expensive SANs and protecting data via RAID to replicating it around so that disks (or whole servers) don't matter.

I run StableBit DrivePool on my microserver at home, which effectively hides all your drives, shows them as one big storage pool and deals with replicating between disks. This is for a few reasons:

- not relying on any built-in RAID, just regular SATA disks to the OS.
- it's all native NTFS disks - if everything went bang I can plug a disk into any windows PC and just get my data
- total flexibility on drive sizes - have gone over the years from 1TB drives up to 4TB ones just replacing one at a time and letting it move the data about.
- replication settings on a per-folder basis. Important docs, family photos? Have lots of copies, on every drive if you like. Backups of VMs that I could reproduce in a few hours? Maybe on two disks, just in case. Library of software ISOs that can be easily re-downloaded? Only keep it on one, I don't really care if they disappear.

As for backing up 15TB, the 8TB archive drives in external format are around £190 now, get a couple and write your backups to them. Ideally get a couple more and keep them somewhere else. But it's really worth having a think about how important different data is to you and how much value you place in backing it up.

[budgie smuggler]

buggalugs said:

Oh I've luck'ed myself out of one or two RAID situations in the past

I think a big thing is to make sure periodic scans are enabled otherwise you can develop bad sectors across the drives and not know about it until you come to read them, i.e. rebuild. If you're scanning monthly then it should be the scans that find failing sectors as they happen.

Also, what does URE mean? I haven't seen it defined anywhere. If you read the same sector again and again does that metric apply? Or is it the chance of any sector going bad in some period of time? Is it actually the write that doesn't work properly leading to the read failiure? Or is it a read head signal thing that means that it's just got to wait a few ms for the disc to spin round and try the sector again? The stat's meaning it just taken at an assumed face value. Not to mention that it hasn't changed much since sata was invented which makes me suspect that its' a figure someone plucked out of the air once.

Also why doesn't this apply to RAID 1 - if you're supposed to average an 'unrecoverable error' for every 12TB read (really??) why wouldn't a 12TB RAID10 rebuild fail just as often? Within an order or magnitude anyway.

Anyway that's just my thoughts on it, I'm not going to go all nerdrage on it it just seems like a storm in a teacup to me.

IMO it's good to be skeptical of these things

AFAIK, URE is essentially just a modern name for a 'bad sector' i.e. one which cannot be recovered within the allowed amount of retries (depending how TLER is set etc)

You're right about the 12TB disks, you would hit the same issue, it's just that 12TB disks are new and I'd forgotten about them TBH.

The 10^14 figure comes from the data sheet for WD Reds. Our HP ent drives are 10^15, so a good deal better.

ETA, the spec says 'better than 1 in 10^14'

Edited by budgie smuggler on Wednesday 14th June 10:40

[budgie smuggler]

sjg said:

I've lived through enough RAID failures at employers and clients to not want that at home. That was proper enterprise controllers and disks too - not huge SATA disks and weak consumer controllers. In the space I work in there's been a big shift away from expensive SANs and protecting data via RAID to replicating it around so that disks (or whole servers) don't matter.

I run StableBit DrivePool on my microserver at home, which effectively hides all your drives, shows them as one big storage pool and deals with replicating between disks. This is for a few reasons:

- not relying on any built-in RAID, just regular SATA disks to the OS.
- it's all native NTFS disks - if everything went bang I can plug a disk into any windows PC and just get my data
- total flexibility on drive sizes - have gone over the years from 1TB drives up to 4TB ones just replacing one at a time and letting it move the data about.
- replication settings on a per-folder basis. Important docs, family photos? Have lots of copies, on every drive if you like. Backups of VMs that I could reproduce in a few hours? Maybe on two disks, just in case. Library of software ISOs that can be easily re-downloaded? Only keep it on one, I don't really care if they disappear.

Nice, I use SnapRAID in a similar way but that sounds a lot more fine grained.

[Slushbox]

sjg said:

I've lived through enough RAID failures at employers and clients to not want that at home. That was proper enterprise controllers and disks too - not huge SATA disks and weak consumer controllers. In the space I work in there's been a big shift away from expensive SANs and protecting data via RAID to replicating it around so that disks (or whole servers) don't matter.

<snip>

I

Same here, rather than blow big money on a single point of failure, we're spreading the essential backups (reprographics - mostly image files and InDesign) around workstations.

Last two days, had to decide whether to bin a 2010 Xeon Dell Workstation or revive/replace it. Figured that spending £2.5k on a new box was pointless, and with the discussion over WD Red or Seagate Barracuda on here, picked up two Seagate Barracuda 1TB drives. 7200 rpm Barracudas are fast enough in RAID 1, it turns out.

Seagate warranty is two years, WD, three, if that's a factor.

[buggalugs]

Red Pro's are nice too - more cache, 7200RPM and 5 year warranty (vs 5400 / 3 year for normal Red's)

Bit on the pricey side though.

[Angrybiker]

TonyRPH said:

How many of you have a 36TB array at home?

Only 20 here

[Murph7355]

buggalugs said:

...
Also why doesn't this apply to RAID 1 ....

It applies to any disk in any situation. It's just an error rate. And any decent controller should be able to handle errors like this.

My NAS units actively monitor the various SMART indicators on the disks installed and email me when things change. I can then make a decision on replacing them. Which I have done on numerous occasions without issues.

As a little bit of irony/karma, one of my NAS units went pop yesterday! Looks like the power supply is toast. The unit's 11yrs old and been in a relatively harsh environment so I can't be too upset.

Moved the disks to another chassis and they all spun up nicely. One disk reported a reallocated sector. Data's now transferred onto my other units while I decide what to replace the dead chassis with (almost certainly another ReadyNAS).

[budgie smuggler]

Murph7355 said:

And any decent controller should be able to handle errors like this.

That's just it though, during a rebuild they cannot.
During a scrub, yes it may be recoverable without losing the drive.

[Murph7355]

budgie smuggler said:

That's just it though, during a rebuild they cannot.
During a scrub, yes it may be recoverable without losing the drive.

Not IME.

[Sam.]

Slushbox said:

The WD Reds have a few extra features over WD Greens; TLER support, useful in RAID Arrays, a listed MTBF of 1,000,000 hours, plus a 3 year warranty. They also have a longer head parking delay and 3D Active Balance (vibration damping.)

However we have Greens in our NAS boxes with no issues (Yet.) We also have a Seagate Barracuda 500GB running in a Dell Precision workstation since 2010. It has no SMART errors or other problems, so it's become a curiosity, like those light old bulbs that never burn out.

The question of RAID 1 over no RAID is harder to answer, RAID 1 doubles your storage cost, (halves the total capacity) but most NAS boxes will re-sync a RAID 1 pair if a single drive is replaced.

So, Reds is good, SSD's tend not to show much improvement, as the transfer speeds from many NAS boxes are fairly low, even over wired Gigabit Ethernet (10 MB/sec on some of them.)

If it's a critical NAS, then adding an Uninterruptible Power Supply (UPS) feed is a cheap way to keep it happy.

Edited by Slushbox on Monday 12th June 07:03

Been running 8 x 4TB WD Blues in my CentOS server for over 4 years now. No failure of drives. Had around 1 hour downtime since it was booted up.

Cant fault them for the price. Im really surprised they havent failed yet