Mushkin shows off a 4TB SSD with estimated price of US$500

[Axle_Grease]

There's something very wrong with your figures there - a typical sata 7.2k disk will give you something like 75iops, so putting 3 in RAID0 (ie no RAID tax) should give you something like 225 iops.

What util are you using to measure iops?

[scaryjim]

What util are you using to measure iops?

When drives quote iops performance it's for random access. You're using 0% random i.e. you're measuring sequential performance, which spinning rust isn't too bad at.

Hexus review using iometer to measure random IOPS: http://hexus.net/tech/reviews/storag...-480gb/?page=6

The high end drives get ~ 70k, mid-range vary from 35k - 50k, entry level Crucial m4/OCZ Octane drives get ~ 10k, 1TB Samsung HD gets 162.

[Unique]

this would be perfect for my laptop setup as currently 2tb drives are the biggest that will fit, unless I did some crazy jiggery pokery with one of the 4tb 2.5" drives to get it to fit

[kingpotnoodle]

I did some tests on my PCs using Iometer (4KiB 100% Read; 0% Random)

Samsung 850 Pro SSD (SATA 6Gb) 73K IOPS (299MBPS)
3 x 600GB WD Black Caviar HDD (RAID 0) 42K IOPS (172MBPS)
Seagate 512GB HDD does 23K IOPS (96MBPS)

So that 10K IOPS figure has got to be a mistake, otherwise it would mean that the Mushkin SSD has less than half the performance of a regular HDD.

The 850 Pro result is vaguely right but those HDD results are way off, perhaps the test data is ending up in OS caches rather than hitting the disk, lose the K from the numbers and it might be closer to correct!

[Axle_Grease]

When drives quote iops performance it's for random access. You're using 0% random i.e. you're measuring sequential performance, which spinning rust isn't too bad at.

Hexus review using iometer to measure random IOPS: http://hexus.net/tech/reviews/storag...-480gb/?page=6

The high end drives get ~ 70k, mid-range vary from 35k - 50k, entry level Crucial m4/OCZ Octane drives get ~ 10k, 1TB Samsung HD gets 162.

Iometer (4KiB 100% Read; 100% Random)

Samsung 850 Pro SSD (SATA 6Gb) 57K IOPS (224MBPS)
3 x 600GB WD Black Caviar HDD (RAID 0) 703 IOPS (2.88MBPS)
Seagate 512GB HDD does 276 IOPS (1.13MBPS)


That makes a bit of a difference, doesn't it? lol. Thanks, scaryjim.

[Splash]

Iometer (4KiB 100% Read; 100% Random)

Samsung 850 Pro SSD (SATA 6Gb) 57K IOPS (224MBPS)
3 x 600GB WD Black Caviar HDD (RAID 0) 703 IOPS (2.88MBPS)
Seagate 512GB HDD does 276 IOPS (1.13MBPS)


That makes a bit of a difference, doesn't it? lol. Thanks, scaryjim.

That's more like what I'd expect to see (though still higher than I would expect, to the extent that I suspect you may be seeing caching in action - the trick is that your dataset *must* be larger than the cache on your controller/in your OS)!

[jigger]

Would the fact this is already a pair of drives in an array cause problems is you want to set up a few of these in RAID?

[lumireleon]

OOOOH some people got it wrong.... The fastest enterprise category single rust drive (HDD) spinning at 15,000 RPM cannot manage 1,000 IOPS (1K IOPS) either read or right. A Samsung 850 SSD can manage 100,000 IOPS (10K IOPS), an INTEL 750 series SSD NVe can do 440,000 IOPS read and 290,000 IOPS write.

[badass]

"Only 10k IOPS"

I've seen Dell Equallogics full of 15k RPM enterprise drives doing 5k IOPS and serving a ton of servers!

Otherwise known as slow, ****e enterprise storage.

[SlayerMatt]

While overkill for me, looking forward to the prices of smaller sized SSDs dropping or possibly even this causing spinners to drop. Could do with another couple tb's of storage and a backup drive

[shaithis]

Otherwise known as slow, ****e enterprise storage.

Regardless, if you can run a business full of servers from 5k IOPs, people shouldn't have issues with a 10K IOPs SSD....to put it in perspective for those that do not understand IOPs

[scaryjim]

...to put it in perspective ...

It's something a lot of people find hard to get their heads round, and one reason that logarithmic scales are useful. Going from a HDD to one of these would only be a ~ 9k IOPS increase, whereas getting a top end SSD would be another 90k IOPS again.

But when you look at the orders of magnitude, you're talking about a ~500x increase, followed by a ~10x increase. It makes sense that making that first 500x faster would be much more noticeable than the extra 10x: i.e. getting a slow SSD v a HDD makes a far bigger difference than getting a fast SSD vs a slow SSD.

[BadHead]

This is a great buy for my two data drives which are currently to big and relatively slow WD mechanical drives. This year could finally be the death of the HDD. Good ridance, too.

[xslavic]

I need one please "arrange"a competition.

[abaxas]

I thought all these SSD iop figures were proven to be bogus. Burst yes but are not sustainable as the drives 'cache' writes and then do garbage collection/etc later. A bit like a shingled HD.

What storage you choose should be based on requirements rather than pure numbers. Especially when spinning disks are much easier to provision for in terms of IO.

[Ironbuket]

So if this uses onboard raid, then theres no TRIM support?