Question RAID for speed redundancy

[Igo69]

How fast is RAID for speed?

I want to get a very high capacity HDD and was thinking i should get 2 for RAID 0 speed. It is cheaper than 1 8TB SSD but how fast RAID HDDs and how fast compared to 1 SSD?

 

[bigboxes]

Not even close. Get an SSD for your OS boot drive. You don't need an 8TB one unless you've got thousands of dollars to burn. Just get a 2TB (or less) for your boot drive. Then get an 8TB HDD for long term storage, if that's what you need.

 

[Insert_Nickname]

I want to get a very high capacity HDD and was thinking i should get 2 for RAID 0 speed. It is cheaper than 1 8TB SSD but how fast RAID HDDs and how fast compared to 1 SSD?

It's not even funny. At best you may be able to pull 300-400 IOPS out of a two drive 7200RPM RAID0 array. Even the most low performance SATA SSD you can imagine can do 10.000 IOPS. A class A1 SD card is faster, except in sequential transfers. Even simple PCIe NVMe drives do 100.000 easily.

In short, there is absolutely no reason what-so-ever for using RAID0 today. A boot SSD, and a large capacity drive is the way to go. Don't forget a backup solution for important data.

 

[Igo69]

Amusing, i never mention anything about boots.

HDDs is for storage media and I need fast reads hence about RAID question since SSDs above 4tb has absurd prices.

 

[kschendel]

You can't get around the mechanical latencies of hard drives. Striping (RAID 0) helps a little, especially if you have multiple processes doing I/O. You'll never get anywhere near SSD speeds, and you double your chances of failure. The flip side of course is that as you observed, large capacity SSD's get very expensive. Cost vs speed is your choice and there's no magic wand that gets you SSD speeds at low cost.

If the majority of the read requests come from a single process / thread, and if they are largely sequential, I doubt that RAID 0 will gain you much; you might as well use a single drive. If the reads are mostly random or are coming from multiple processes, RAID 0 helps hide some of the seek / rotational latencies. Don't expect anywhere near 2x improvement though.

 

[bigboxes]

Amusing, i never mention anything about boots.

HDDs is for storage media and I need fast reads hence about RAID question since SSDs above 4tb has absurd prices.

You never mentioned a usage scenario. You still don't. No on the RAID 0. Just another point of failure with little gains. Get what you can afford. I've got a second 2TB NVMe drive for games. I've got four spinners as thrash drives. If you need the large capacity drives for your work then you need to do a cost/benefit analysis to determine if it's worth it to have faster load times. If your employer is paying for it, then 8TB SSD it is.

 

[Igo69]

HDD is for storing videos and photography.

Can you explain what is failure in RAID because 1 HDD can also fail or failures in RAID are different and occur more often than in simple HDDs.

What is so bad about RAID 0?

Not looking into cost/benefit. I need a hard drive with fast reads. I can go and buy a 15TB SSD that cost $3600 (LOL) but will never do that since i think the price is (again) absurd.

I guess i have no choice and just buy a single HDD since RAID is so bad.

 

[Billy Tallis]

Can you explain what is failure in RAID because 1 HDD can also fail or failures in RAID are different and occur more often than in simple HDDs.

In a two-drive RAID-0 array, the probability of drive A failing is mostly independent of drive B failing. Either one of the drives failing kills the array. So the probability of the array as a whole failing is approximately twice the probability of a single drive failing, and this trend continues as you add more drives to the array. RAID-0 isn't just a way of aggregating drive capacity, it's also a way of sacrificing reliability to gain sequential throughput.

Not looking into cost/benefit. [...] but will never do that since i think the price is (again) absurd.

Contradicting yourself within the same paragraph is not conducive to a productive discussion. In order to make an informed decision about what storage setup makes sense, we need to consider:

• approximate capacity requirement
• intended use case, so that we know which performance metrics matter and which ones don't
• approximate budget
• maybe something about your reliability requirements, if they're unusual

On the performance front, you've said media storage, but have not quantified "fast reads". Media playback—even to multiple clients—is rather different from something like a nonlinear editing system.

 

[gdansk]

RAID0 is not recommended. One disk fails, there goes your data.
Those drives look good and have a good warranty but comparing to SSD it will be worse if you care about speed. Even with 2 of them in RAID0 you'd probably only get 500MB/s sequential read with a much higher probability for data lost. Almost all NVMe SSDs can get 5-10x that in sequential read scenarios... for speed there really is no comparison.

I do have 4 of those 14TB Toshiba drives set up in RAIDZ1, but I am under no illusions as to their speed. It's for when you really do not need speed

 

[Fallen Kell]

RAID0 is not recommended. One disk fails, there goes your data.
Those drives look good and have a good warranty but comparing to SSD it will be worse if you care about speed. Even with 2 of them in RAID0 you'd probably only get 500MB/s sequential read with a much higher probability for data lost. Almost all NVMe SSDs can get 5-10x that in sequential read scenarios... for speed there really is no comparison.

I do have 4 of those 14TB Toshiba drives set up in RAIDZ1, but I am under no illusions as to their speed. It's for when you really do not need speed

They work pretty well, but I have 6 of those in a raidz2 (with 192GB RAM and a Optane ZIL).

 

[kschendel]

The RAID0 / striping reliability issue doesn't need to be a showstopper. Striping can be a useful performance tool as long as your expectations are appropriate (ie it's not magic). You just need to be aware of the situation and adjust your data placement and backup procedures accordingly. If you have critical data on a striped volume, you either need lots of backup, or you need to consider e.g. striped mirrors instead of striped bare drives. That of course costs more money, no free lunches.

If you're storing videos and photos, that sounds kind of important, and striping isn't really going to buy you all that much as I'd expect many read operations will be sequential. I'd put the money into mirroring the volume rather than striping it. (Besides, a properly implemented mirror can modestly improve read access times in some cases by having both drives serve portions of the request.)

 

[00Logic]

To actually answer your question:

RAID 0 will increase large sequential speed
but
It adds another layer, to be processed, to the I/O stack which adds latency and that generally isn't good for Small Random (4K) I/O.

R4K at Que Depth 1 is very important to the Windows OS as:
66% of all I/O is R4K and QD1,
NOT:
The large sequential #s advertisers have waved around like burning flags untill the world believes them most important!
Less than 1% of Windows I/O is large sequential..!

Understanding SSD Advertised Performance and Its Purchase Implications – An SSD Primer - The SSD Review

Understanding SSDs Performance and Implications Today's SSD close up is going to teach us the most v

www.thessdreview.com

Newer games are doing more 16KB and 32KB reads:

How Many Files Do Games Load? Does 4K Performance Matter to SSDs? | GamersNexus

stub SSD benchmarks generally include two fundamental file I/O tests: Sequential and 4K random R/W. At a very top-level, sequential tests consist of large, individual files transfers (think: media files), which is more indicative of media consumption and large file rendering / compilation. 4K...

www.gamersnexus.net

(But also NB; lots of 4KB writes)

Because HDDs have a very high seek time/latency vs SSDs the added I/O stack latency isn't generally a problem as the CPU has a lot of time 'to think' during these seek times.

As a HDD rule of thumb:
You get double the R4K write speed, but slightly lower R4K reads. with RAID0.
Larger stripe sizes improve Small Random I/O at the expense of Large sequential.

NB:
That no HDD RAID arrangement can come close to competing with the random I/O speed of a SSD.

A compromise:
If you tend to play the same game for a while before changing to another, I recommend SSD caching:

Free:
- Intel: Rapid Storage Technology (RST)
- AMD: StoreMI
- ReadyBoost:
Despite all the BS on the net, RB is simply the caching of R4k to a media that can read 10+ R4K files in the time a HDD takes just to seek 1...
All while the HDD/RAID gets on with the large sequential I/O its good at... simultaneously..!
* Yes SSDs work and the cache will survive suspends. (but not reboots)
(Same for Flash drives IF they are seen as internal. There are filter drivers to do this if the drive doesn't already)
* 32GB of space can be used per RB drive IF its formatted exFAT or NTFS.
(That's 8 000 000 R4K files. More 'all of them than a cache..)

There are some other tricks to speed up RB if you're interested:

That's a lil ol 2.5" notebook HDD

Best:
Primocache:
L2 cache on a SSD partition. (and add a delayed write L1 cache in DRAM)


eBoostr:
A file aware cache where you can specify which apps/games/files/folders to prioritize or ignore.
NB: that this cache is updated once an hour (unless you do so manually) so ignore any bitching that "It dont work!" (after 5 minutes) that you see on the net.
Yes it works on W in 10.