In spinning rust we TRUST: HGST slips out screamingly fast ... HDD Western Digital sub HGST has decided there's life in small fast pure spinning disks yet, and launched a 600GB, 15,000 rpm drive for enterprise data centres: the Ultrastar C15K600. Is this the last hurrah of the fast and pure disk drive? This is not a hybrid drive, one with a slug of caching flash, such as is used by Seagate's …
You shouldn't be able to hear any rotational noise at all - if it's whining, it's dying!
What you hear much more on datacenter disks than on home ones is head seek noise. There's a tradeoff between minimising acoustic noise and minimising head repositioning time. In a datacentre, noise usually doesn't matter (it's dominated by lots and lots of fans moving air around, often making it hard to be heard without shouting).
It's likely to make a bit of noise, but not too much. I have an old SGI machine at home which has a 15K 3.5" system disk and although it's audible, the noise is surpisingly low. And as said, noise in a datacenter is less of an issue anyway due to cooling fans, aircon and the likes.
Noise! Ha! I have, safely salted away in the back room, a pair of Seagate Cheetah X15 drives, the first-ever 15k units (which are not too bad noise-wise, not considering how old they are) and somewhere near them, a couple of older Seagate Cheetah Mark 1 drives, the very first drives to spin at 10,000 RPM. (Everything prior to that was 7200 max.) The Mark Is were very, very fast (by 1997 standards, or the standard of several years later) but very loud. Quite unpleasant to be in a small room with one after a while.
But they were nothing, noise-wise, to the IBM model which followed soon after: the Ultrastar ZX - the second 10k drive model to be sold, and possibly it was a bit rushed to keep up with Seagate because it made a noise like a small jet fighter taking off. Really loud; a penetrating note that set your teeth on edge and made you wonder if it was quite right in the bearings. But it was huge (9.1GB!) and fast, and it ran without the slightest trouble in our office server for six or seven years, 24 hours a day. It still runs now - not that I switch it on more than once every couple of years just to hear that rushing mechanical whine again, and maybe watch the streetlights dim as it powers up - and provides performance vastly inferior to a $10 memory stick from the Post Office with 10 or 100 times the capacity and no noise at all. But where is the glorious mechanical engineering in a memory stick?
Ahh ... push me off the perch, I'm getting old.
Should also last longer than SSD
Depends on how it's being used. In a write-mostly and intensely-accessed environment an SSD will "wear out" in less than the several years that a mechanical drive can be relied on. In one where reads are more frequent or where there is 16 hours/day near-idle, there may be 10, 15, more years of write traffic needed to wear out the SSD, and in that case I'd expect the MTBF of an SSD to be higher than an HD. Expect, because SSDs haven't existed for long enough to be certain about their long-term ageing characteristics.
Given it is a SAS drive and HGST's 600GB 10,000rpm drive is ~200GBP, then a lot lot cheaper than the equivalent 400GB or 800GB SAS 2.5" drive which for 400GB is north of 1000GBP at the moment.
Yeah SATA drives are cheaper but you don't get a dual ported interface or Data Integrity Field (DIF) 520 byte sector sizes. If you don't know why either of these might be important you are not qualified to comment.
Yeah SATA drives are cheaper but you don't get a dual ported interface or Data Integrity Field (DIF) 520 byte sector sizes. If you don't know why either of these might be important you are not qualified to comment.
It helps not one iota if the entire unit is bricked. I also think you'd be surprised just how much error checking and recovery is built in to a standard hard drive even before RAIDing them - you never see the vast majority of read errors. In any case I don't think it's particularly relevant since a lot of data storage bods are naturally very conservative and want to see evidence of long term reliability over a couple of generations. SSDs have been around just about long enough to do that now but the long term figures from four or five year old units are far from encouraging.
Offer them the odd millisecond based on established and trusted technology and they'll take it. Offer them five milliseconds based on technology with an appalling reliability record less than a generation ago and they are much more circumspect. Anecdotal evidence of the "I've had an SSD in my desktop for eighteen months now and haven't had a hitch" is not going to sway them from that position.
How about this for anecdotal evidence?
1) I've brutally punished my Intel 520 480GB SSDs for over a year without a hitch. By brutally punished I mean "ran IOmeter and SQL bench on the things in every configuration imaginable." Because it's my job to do so.
2) I've been testing-to-destruct with a Micron P420m PCI-E SSD for about 6 months now and the thing isn't even past 1% write lifetime used. As near as I can tell it's actually made out of indestructible.
3) I've run my Kingston Hyper-X 3K 240GB SSDs (8 disks in RAID 5) in production (they support about 50 VMs ranging from SQL servers to VDI) for over a year now and they have proven themselves to be entirely reliable. They are at approximately 10% of write life used.
I've had consumer SSDs in production for about three years, no real grief. Some dead disks, but no more than mechanical drives. Enterprise SSDs do a thing consumer ones don't: when the enterprise SSD turns into a pumpkin it goes into read-only mode, which lets you get your data off. Consumer ones just die.
As you say, hyper-conservative types will need more time in the field before they trust the technology, as they don't trust anything until after the entire rest of the industry has moved on to something else entirely. But to say that the only evidence for trusting SSDs at this point is "it runs good on my home system" is bunkum.
These things are in real servers, in production...and not giving any more crap than mechanical drives.
How about this for anecdotal evidence?
1) I ...
2) I ...
3) I ...
How is that not anecdotal?
How does it contradict our first stab at in-house testing? In 2008 we installed 12 32GB drives in desktops from two different manufacturers. After two years 50% had failed. After four 75% had failed. 100% failure was at 4 years 8 months. We are conducting a second experiment as we speak on admin workstations but in the first twelve months sings are not encouraging.
Our DBA's job is to protect the data at all costs. You don't have to convince me, you have to convince him.
First off, I said it was anecdotal. Secondly, I listed specific manufacturers which had done me well. others - such as OCZ - have not. In fact, I have now a 240% failure rate with OCZ. (Of a few hundred units in the field.) That's right, I have had so many OCZ drives fail that the RMA replacements in some cases unto the fourth replacement have failed.
I am also using the SSDs I quoted in situations far more punishing than any desktop. I am using them in server workloads - including supporting multiple production databases - in RAID arrays. If you know a bent damn about disks you know that using them in RAID brings layers of additional complication (such as array rejection, timeouts, rebuild, etc) that go beyond the lifespan of an individual disk.
So yes, my evidence is entirely anecdotal, but it provides you specific models from specific manufacturers as well as an idea of workload and provisioning arrangement. That should be enough to start looking for corroborating evidence from others to determine if the models and manufacturers in question are trustworthy. (Hint: Micron and Intel absolutely are. They share a joint fab and make the best flash in the industry.)
You are basically writing off an entire technology because of some bad desktop trials of what I am assuming are consumer drives. From the sounds of it, no very good ones at that.
There is a world of difference between consumer SSDs and enterprise drives. eMLC is a hell of a lot more resilient than standard MLC and SLC is even better still. If you - or your DBA - are so prejudiced against a technology that you will grasp at any negatives possible for an excuse not to use it, then go hard and have fun.
To be blunt about it: your irrational prejudice means that there is more of a precious resource available for the rest of us. You go, fret about your inability to keep SSDs working in workstations. I'll run them in my servers and I'm never looking back.
It does, however, strike me that not all anecdotal evidence is equal. Nor are all trials or tests. Bear one thing in mind about all of this: my personal economic incentive is to find a problem with products.
If I could prove that even one model from one manufacturer was conclusive shit - let's put OCZ to one side because everyone knows the vertexes are complete shit - then I get to write an exclusive expose and put a nice fine feather in my cap. Writing articles that say "it does what it says on the tin" aren't exactly exciting, nor prestigious. So I go out of my way to find problems; I look for corner cases and I test things beyond the redline wherever possible. Hell, IOsafe wanted me to test one of their NASes so I lit it on fire.
This is what I do for a living. So it could be that maybe - just maybe - if I can't break the damned things, then not all of them out there are shit. It's all anecdotal, of course. It's not like there are entire multi-billion dollar industries pretty much running on flash (all flash, tiered or hybrid) which could serve as additional case studies to back up my lab results.
Oh wait...
Twice as fast as previous disks? Perhaps the 12Gb SAS can go twice as fast as 6Gb SAS but this disk has similar sustained performance to older disks. That said, the SAS loop is becoming quite the bottleneck in SAN systems these days so it's good to see an improvement even if the disk isn't able to get much benefit from it.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
