1–1011–19

Silly Question

From: Ken (SHIELDSIT)22 Dec 2015 02:44
To: ALL1 of 19
41647.1
Can someone here do the maths (good God, what have I become?) and tell me how much worse a 2 drive RAID-0 array is vs separate dual drives? I almost had myself talked into it being no different, but I now realize that's not true, but I don't know how much difference there is!

Thanks smart peeps.

** Here's a softball for you PB!
Reply
0/0
From: CHYRON (DSMITHHFX)22 Dec 2015 08:00
To: Ken (SHIELDSIT) 2 of 19
41647.2 In reply to 41647.1
Define "worse". RAID 0 is supposed to increase data throughput. YMMV, depending on various factors.

https://en.m.wikipedia.org/wiki/Standard_RAID_levels#Performance
EDITED: 22 Dec 2015 08:02 by DSMITHHFX
Reply
0/0
From: Ken (SHIELDSIT)22 Dec 2015 08:50
To: CHYRON (DSMITHHFX) 3 of 19
41647.3 In reply to 41647.2
Without rereading what I wrote, I mean failure rates.  How much faster will a raid 0 array fail over a single drive failing?  Is it 50% faster due to doubling the drives?
Reply
0/0
From: CHYRON (DSMITHHFX)22 Dec 2015 11:34
To: Ken (SHIELDSIT) 4 of 19
41647.4 In reply to 41647.3
I don't believe the failure rate of the hardware is increased, but the risk of data loss is doubled (failure of either drive = total data loss).
Reply
0/0
From: Ken (SHIELDSIT)22 Dec 2015 11:55
To: CHYRON (DSMITHHFX) 5 of 19
41647.5 In reply to 41647.4
Correct, loss of data is what I was meaning.  So there is no funky formula? It's just a straight up 50% increase?
Reply
0/0
From: Chris (CHRISSS)22 Dec 2015 13:08
To: Ken (SHIELDSIT) 6 of 19
41647.6 In reply to 41647.5
Does this help?
Attachments:
Reply
0/0
From: CHYRON (DSMITHHFX)22 Dec 2015 14:38
To: Chris (CHRISSS) 7 of 19
41647.7 In reply to 41647.6
I was going to say that, but I didn't want to take Ken away from the drone snowplow project.
Reply
0/0
From: CHYRON (DSMITHHFX)22 Dec 2015 14:50
To: Ken (SHIELDSIT) 8 of 19
41647.8 In reply to 41647.5
Probably more highly influenced by brand/model/age of drive(s) than anything else. Let's say you have a brand-new hdd with a statistical 4% failure rate (I think this is fairly common among consumer-grade hdds -- the rate goes up with age). If you have another 'identical' drive coupled with the first in RAID 0, you now have a statistical failure rate of 8%. Complicated by the fact that it is much easier to recover some or all data from a failing single drive, than from a failed RAID 0.

If you want some of the performance boost of RAID 0 with the redundancy of RAID 1, use RAID 1+0 (aka RAID 10), at the expense of ½ the total hdd capacity.

IANA mathematician, this is just some half-remembered shit I read somewhere a year ago.
EDITED: 22 Dec 2015 14:52 by DSMITHHFX
Reply
0/0
From: graphitone22 Dec 2015 16:40
To: CHYRON (DSMITHHFX) 9 of 19
41647.9 In reply to 41647.8
Quote: 
If you want some of the performance boost of RAID 0 with the redundancy of RAID 1, use RAID 1+0 (aka RAID 10), at the expense of ½ the total hdd capacity

Aye, a mirrored stripe. Does that mean 4 drives, or is there a more efficient way of doing things?
 

Reply
0/0
From: CHYRON (DSMITHHFX)22 Dec 2015 17:08
To: graphitone 10 of 19
41647.10 In reply to 41647.9
Apparently it can be done with only two:

https://en.wikipedia.org/wiki/Nested_RAID_levels#RAID_10_.28RAID_1.2B0.29
Reply
0/0
From: Ken (SHIELDSIT)23 Dec 2015 03:58
To: ALL11 of 19
41647.11
I knew there was fucking goofy math involved with this.  I'll shamelessly steal this from Tom's Hardware and assume the people who chose this as the correct answer are smarter than I am, not hard to be...
 
Quote: 
It is more dangerous than a single drive. But as-per the people before me, not by a huge amount. The odds of a failure are 1−(1−r)^n (Thanks Wikipedia) Where r is the failure rate, and n is the number of drives.

That is, if a drive has a 1 in 100 chance of failing, 1 drive: 1-(.99)^1 = 1% chance of failing, 2 drives: 1-(.99)^2 = ~2%
3 : ~3%

What if the failure rate were higher? I don't know the odds of a given disk failing, ask Google for those numbers. 1 : 5%, 2: 9.75%, 3: ~14.3%

My math could be wrong, but I blame Wikipedia. It seems right to me. As you can see, you really start gambling when you get to higher storage levels. If you are alright with risking losing everything (other than what is already backed up) I say go for it. Personally, I am all HDD, no RAID. Don't have the money to spend on SSDs, especially since I just don't need that performance yet.
 


 
Reply
0/0
From: Peter (BOUGHTONP)23 Dec 2015 12:37
To: Ken (SHIELDSIT) 12 of 19
41647.12 In reply to 41647.1
I do RAID-Z :<
EDITED: 23 Dec 2015 12:39 by BOUGHTONP
Reply
0/0
From: Ken (SHIELDSIT)23 Dec 2015 14:51
To: Peter (BOUGHTONP) 13 of 19
41647.13 In reply to 41647.12
So do I on a few systems, like my file servers, but you can do the math!
Reply
0/0
From: Peter (BOUGHTONP)23 Dec 2015 16:20
To: Ken (SHIELDSIT) 14 of 19
41647.14 In reply to 41647.13
You already did the maths for me?
Reply
0/0
From: Ken (SHIELDSIT)23 Dec 2015 17:53
To: Peter (BOUGHTONP) 15 of 19
41647.15 In reply to 41647.14
Do you agree with that formula?
Reply
0/0
From: Peter (BOUGHTONP)24 Dec 2015 00:14
To: Ken (SHIELDSIT) 16 of 19
41647.16 In reply to 41647.15
If disk failure is relevant, RAID-0 is the wrong answer.

If you're asking from academic interest, you'd need to consider whether striping across disks increases or decreases the likelihood of failure due to different read/write behaviour. (I don't know what the contributing factors to a disk failing are - i.e. is it purely random, actively increased by use, etc.) That formula doesn't include any such parameters; it may well be an approximation of something that does. I don't see on the linked Wikipedia page where the poster apparently sourced it from, so it's currently just a way to produce a number slightly less than directly multiplying the failure rate by number of drives.

Reply
0/0
From: CHYRON (DSMITHHFX)24 Dec 2015 00:30
To: Peter (BOUGHTONP) 17 of 19
41647.17 In reply to 41647.16
Disks can fail in several ways: bearings, actuators, surfàce, PCB. The failure rate increased dramatically after the flooding in Thailand destroyed many mfg facilities, leading to speculation that flood-damaged parts continued to be used. Outside of production servers, the main factors influencing failure rates appears to be make, model and even batch. Some guy running a cloud service periodically publishes failure stats on assorted makes and models.
Reply
0/0
From: ANT_THOMAS24 Dec 2015 10:17
To: CHYRON (DSMITHHFX) 18 of 19
41647.18 In reply to 41647.17
Indeed they do

https://www.backblaze.com/blog/hard-drive-reliability-q3-2015/
Reply
0/0
From: Dave!!28 Dec 2015 12:31
To: Ken (SHIELDSIT) 19 of 19
41647.19 In reply to 41647.1
Another important question these days is the capabilities of the drives and the controller (most controllers do share the bandwidth to some degree between the SATA ports), and why using solid state drives isn't an option. For example, striping mechanical hard drives will almost always still provide poorer performance than a single SSD. Especially seeing as modern SSDs can easily saturate a 6Gb SATA link. For ultimate performance, you'd connect a fast SSD to either a SATA Express port, or a bespoke PCIe controller.

I just can't help but think that RAID-0 is a lot more redundant these days than it used to be. Faster performance, increased risk of data loss, 2 drives required, or just slap a fast SSD in instead. Unless of course you need lots and lots of storage, and even then you can now get 1TB SSDs for less than £60...
EDITED: 28 Dec 2015 12:31 by DAVE!!
Reply
0/0