Is RAID Relevant Anymore

[OreoCookie]

You would like never need an 8TB SSD for still photos. You only need fast access to thrones you just shoot and are working on. a 500GB SSD would work, then you move the processed data to a RAID archive.

I try to be careful with the word “need”, because usually having large SSDs opens up a lot of new possibilities. For example, you an build 3.5" 60 TB SSDs that merely consumes at most 15 W, so you can achieve much higher data densities. And on a price-per-gig level of 0.67 $ per GB, it isn't expensive — although you'd apparently still need to shell out $40k …

As with all things, we are limited by money and availability, and they tell us which compromise works best if we have to balance performance, capacity, price and availability.

 

[v3rlon]

All right, mean time between fails for a decent SSD is north of 20 years. How much redundancy do you really need?

One set of photos locally, one backed up off site, one on time machine. If Aliens land in your driveway, fire off an EMP, and initiate the zombie apocalypse, your screwed way worse than your photo backups. Priorities, you know? Honestly, I think that scenario is a bit on the unlikely side.

A lot of people will tell you "if it isn't in three places, it isn't secure." How many hard drives have you had fail, EVER? Where you were sure it was the drive that went out. Even at work where we have hundreds in my department, we have had maybe 5 MAYBE go out in 20 years or 24-7 usage. And some of these drives are 20-25 years old now.

I am not saying do not back up at all, but needing >3 copies of everything is beyond paranoia and into wasting money.

SSD
Onsite backup
Offsite backup

And you would have to be EXTREMELY unlucky to not be okay.

 

[Ledgem]

I use it more to pool storage while keeping everything centralized. I used to have about seven hard drives all daisy-chained, and keeping track of what data was where was a nuisance. Now I use a Drobo (which essentially functions like a RAID 5 system), and everything is just sorted into folders. The system can grow as I need it, and also has some fault tolerance. Cleared up some desk space, too. I have backups to a cloud-based storage system, just to account for a total failure of the Drobo unit, itself.

Maybe it's a lazy man's usage of RAID, but I've been happy with it.

 

[Chancha]

There is still a clear sweet spot for RAID0 dual HDDs even for consumer/prosumer usages. Since SSDs get exponentially expensive after the 1TB mark, and a typical photographer's archive can easily exceed that, so we will get a pretty good use case in the 4TB-20TB or so capacity in the form of external HDDs. Then the question is how to spread out those data over the disks, and since 6-10TB single platters are now available, RAID0 enclosures of 12-20TB is possible at relatively affordable price, compared to say a 5+ bay array from 3 years ago to get the same space. The benefits of having a single logical volume plus the speed increase from being stripped is welcome for HDDs, if on enterprise grade disks like the G-RAIDs, which uses HGST Helium disks, just 2 disks stripped can already get you about 400mb/s, approaching the speed of a single SATA3 SSD.

I do see a lesser need on the redundancy aspect of RAIDs lately though, since standalone one-disk enclosures are as cheap as they are, and you can never get more redundant than a physically separated unit.

 

[OreoCookie]

I don't think it is a good idea to use a RAID0 in most circumstances. Also the argument that because you are using standalone disk enclosures, you are “as redundant as it gets”, is false. If anything, the risk of data loss is higher, all you need is to lose one of the disk enclosures due to failure, and that is always higher than a single disk enclosure failing (you are increasing the points of failure). Moreover, it means you are relying on software RAID, which is also a riskier proposition.

There are still use cases for which RAID0 is suitable, but you must have a good backup strategy. And you need to be ok with a long recovery time in case of drive or enclosure failure.

With a RAID0, losing one disk still means losing all the data. The main reason why people went with RAID0s was performance and volume size. At least performance is no longer the determining factor, non-RAIDed SSDs are so fast that you are usually limited by how you connect the SSD to your machine. That leaves volume size, but also here, there is less and less need. The largest hard drives you can currently buy are 12 TB in size (and I think a 14 TB drive was just announced). There are fewer and fewer use cases where having “just” 12 instead of 20+ TB of contiguous storage becomes a problem.

 

[Chancha]

Well I never intended to undermine the risks involved with RAID0/stripping (which should be essential knowledge as far as this thread is concerned). And then my statement saying "standalone one-disk enclosure being redundant" is in the context vs RAID arrays with "internal" redundancy. And with that I am implying usage of an extra external disk as a discrete archive, with backup software of your choice or even manually, the unit can even be taken off site / on rotation. In contrast, in-enclosure redundancy offers more immediate safety than long term, thus "not as redundant". Sorry to be too ambiguous earlier, all I wanted to say was that "since single HDD enclosures with somewhat large capacity are cheap anyway, why not get a bunch of them as backup drives, instead of setting up any RAID if redundancy is the concern".

Having a single logical volume for easier media management is always one of the biggest reasons for stripping / concatenate. I myself see the extra read write speed in dual drive RAID0 just a bonus, as it merely doubles off slower HDDs which is still rather slow. Of course the speed increase in multiple bay arrays is much more favorable, but at a cost.

 

[MCAsan]

I see the large enclosures for 4 or more drives. Great the you can run RAID 0 on all them for very fast performance or even RAID 5 for a good compromise in speed and redundancy. But with the enclosure or its power supply dies...everything is offline until a repair or replacement happens. Can not see using just one enclosure. Ideally a RAID 0 in one enclosure for the libraries and a second separate enclosure for the single disk or RAID 1 that runs Time Machine backups.

The point is, also consider the enclosure arrangement, not just the disks inside.

 

[OreoCookie]

But with the enclosure or its power supply dies...everything is offline until a repair or replacement happens. Can not see using just one enclosure.

I understood Chancha as him connecting a bunch of external hard drives in separate enclosures and then using a software RAID0 on them.

You are right that if you have something like a NAS, then the NAS itself becomes a single point of failure (e. g. the power supply or some other part). As with anything in life, you have to balance security, cost and convenience. Personally, I think creating a different volumes on the same NAS is an acceptable risk if you possess other backups of your data.

@Chancha
I still don't think this is a very good solution. A single drive will always be more reliable, and I reckon you will be much better off with a single high-capacity drive and an SSD for fast storage. If you have massive storage needs, get a NAS with at least 4 bays and use RAID5.

Also, concatenating is not the same as RAID0 (sometimes referred to as striping (hard drives don't get undressed)): when you concatenate drives (the mode is called JBOD, just a bunch of drives), you have zero speed benefits, because you first fill one drive and then the second. If one drive dies, you lose the data on just that one drive. With RAID0, half of the data is written to one hard drive, the other half to the other. If one drive does, you lose all data.

 

[QuantumLo0p]

Yes, and I linked to real world test data.

Even consumer-grade SSDs are on par with hard drives under the most favorable conditions for the latter. As soon as you add random seeks to the mix, the SSD will vastly outperform spinning platter hard drives. Especially if you stick to quality SSDs from manufacturers such as Samsung or Intel.

RAIDs are no longer the weapon of choice for performance, it is about volume size and redundancy.

I'm still seeing the older ssd's fall flat on large, sustained writes on all them that I own/use. Samsung 840/850 Pros and also Microns too. The latter is on some test equipment which I do not own; the former I had purchased. The newer ssd tech is all that's worth anything large writes; the old ssd tech, well, is old. Are older ssd's still valuable tech? Sure they are but burst write speeds and consistently fast read speeds are where they shine.

 

[MCAsan]

RAIDs are no longer the weapon of choice for performance, it is about volume size and redundancy.

Depends on the budget. At over $600 each for a 2 TB SSD a 8TB RAID 0 would be $2,400 (4 SSDs) plus the enclosure. A nice G-Tech 8TB Raid 0 (2x4TB) is around $600. That works for me.

SSDs are not yet available in larger sizes like HDD and are not available any anywhere near the same cost per TB.

 

[OreoCookie]

Depends on the budget. At over $600 each for a 2 TB SSD a 8TB RAID 0 would be $2,400 (4 SSDs) plus the enclosure. A nice G-Tech 8TB Raid 0 (2x4TB) is around $600. That works for me.

A single hard drive would be safer, though.

Personally, I don't need 8 TB of contiguous fast storage, 1 TB is enough. Once I am done, I can move my projects off fast storage. Then I don't care that it is a NAS which I connect to via wifi, speed is secondary. That's a better, safer and faster solution for me. YMMV, though.

 

[MCAsan]

It looks like the fastest HDD is the WD Black series with the 6TB model having the fastest effective speed. From tests I saw it has read/write in the 200+ range. That is about equal to my current RAID 0 using Seagate 3TB drives that are around 4 years old. A RAID 0 made from them should have read/write in the 300+ range. While maybe not in SSD territory, it should work well for me.