Why don't UK enterprises use MAID? This is puzzling. We live in an increasingly green era and spinning disk platters around while no one is using them is like having empty lorries drive up and down a motorway. Chris Evans, a UK storage consultant, positions MAID this way: "(It) refers to Massive Array of Idle Disks, which …
Problems with MAID
The analysts are right to point out fear of change but here are some things they have missed;
1) Power is a very small part of the ownership cost of storage as the hardware has yet to commoditise in the same way as servers. As the cost of storage tin falls power consumption will become a larger component of the price and thus more of an issue.
2) MAID add on software is of little value, it does not reduce the power and cooling that needs to be provisioned to the devices and therefore has very little impact on the device cost or energy consumption in the typical corporate data centre. (fixed costs dominate)
3) Copan type MAID solutions which limit the total power through constraining the number of drives spun up at any one time allow operators who are power or space constrained in a data centre to install more TB than other kit, thus the uptake from banks in the city who have run out of power and space.
Why don't we use MAID?
A question that could only be asked from those not actually involved in working in large datacentres managing thousands of servers and many petabytes of storage.
Where you have a large, mixed storage infrastructure, there is always something that wants access to data. Even with nearline requirements, it only needs one request by one server that wants access to the data to require the whole RAID set to be spun up. If your RAID set is made up of a 7+1 RAID5 collection of 1TB drives, that represents 7TB of usable space. That's an awful lot of potential data which only requires one IOP per minute to effectively nullify all the power savings. Each disk has to be spun up (which, itself, uses considerably more energy in a few seconds than, perhaps, keeping the platters spinning for perhaps 30 seconds), do the IOP, wait a while before deciding that there's unlikely to be anything more to do, then going to sleep again.
It only requires a tiny amount of I/O to effectively make MAID just another RAID configuration. In many cases, organisations that do have nearline or semi-archival type requirements have data spread over a very large storage infrastructure and often struggle with the problem that large capacity disk drives have atrocious access density (IOPs per GB) capability. All it requires is one IOP per 7TB per minute and the case for MAID is just ruined. Good luck to you if you have the sort of application where you can put data to sleep for hourse on end - in our case we have PB of data, and you can absolutely guarantee that something, somewhere needs data. Providing you are running your storage arrage to a reasonable throughput utilisation MAID is completely useless.
MAID is also not of much use in disaster recovery or remote mirros if there is a requirement to keep data reasonably up-to-date
You can use MAID for tape substitution, but frankly tapes are cheaper, lighter, more power efficient and more robust for this sort of thing (try shipping MAID sets off-site).
I, like many, backup my PC to an external drive and for small organisations this is a sensible way of handling server contents (which is a sort-of MAID-on-the=cheap scenario) However, for the vast majority of big organisations the stated energy savings for MAID are illusory as they are based on wholly unrealistic workload and data access scenarios. It's possible to come up with examples where it might work, but every time I've looked at it on a real workload the whole principle just breaks.
So don't blame the organisations - try understanding your own assumptions about data centre workloads and compare them with actualities first.
How much power is saved with this, really?
Sure, it takes some power to make a hard disk platter go round - that's beyond dispute. But in the context of a machine easting 400W, every second of the day, it's pretty damn minimal. I'd be more impressed if this marketroid could tell us how much power was being saved. Oh, and how much extra it cost to buy the kit with this option enabled...
SurReal Politically Correct Virtual Coup
..... with Pirate Controls .... Red Hot Handles/Immaculate Keys.
"As a result it's very much to do with the relationship with vendors (who aren't pushing spin-down products)." ..... That will result in spin down/spin off Products Takeover/MakeOver
We've a good Story to Tell, go out and tell IT and IT will Foretell it with Virtual Virgin Media Controls .... MODified Special IntelAIgents Services, Darling.
AI ProgramMING for Government Backing .... Astute Betting .... Advanced IntelAIgent Investment, which are Instant Currency Credit Transfers from the National Debt thus Funding NeuReal Technologies Cash Flows Painting the Future into the Black...... by Simply Presenting IT with MODified Special IntelAIgents Services, Darling.
Which is a Registered Live Transparent Private BetaTest of Governments Virtual Services for dDelivery of Cyber Security Operations Centres on AIMissions/ProgramMING Projects for Virtual Realisations. ..........Leading Presentations with C42 Quantum Communications Systems.
I Kid U Not.
I agree with Steven Jones
I work for a company with many PB of onilne data, MAID systems would be useless for online data for the reason that everything would be spun up all the time. The only place that it could be usefull is for backups, but the cost differential betweem backing up to tape and backing up to disk or via disk is too big. There is also the cooling cost when they are in use, an L5500 or a TS3500 tape library needs next to no cooling and uses little power, the cooling required for a MAID array and the power cost are too large.
Sounds like a solution...
...looking for a problem.
MAID suffers from lack of a real volume management system and a real file system
MAID could be made more successful if it was packaged with a real volume management and file system management system, like open-source ZFS.
- The concern about disks not spinning up could be resolved with RAIDZ2 or RAIDZ3 (where 2 or 3 extra parity disks could be available)
- The concerns about the occasional I/O having to spin up all the drives for odd-writes or keeping D-R locations in-sync could be resolved with a ZFS intent log (with a very large cache.)
- A real file system and volume manager could be tuned to deal with delaying writes to disk (as long as the writes were committed to intent log in NVRAM) for hours, ensuring power savings.
- The concerns about the extra power required to spin up the disks could be resolved with super capacitors to deliver the spikes of power needed with spindowns, smoothing out the need to have huge power supplies (running inefficiently), to deal with the power spikes.
- having huge volumes, which can extend [virtually] indefinitely with ZFS, would make MAID more appealing, since splitting data-sets over [sequentially accessed] tapes and [randomly accessed] standard file systems would go away, making the media more appealing in large scale deployments
- pushing the logic to spinning down the drives into the file system / volume manager like ZFS with MAID could make the technology go mainstream, since the capabilities would not require human intervention (with the exception of specifying whether the MAID caching algorithm would need to be mild, medium, or aggressive) - but this would not do too well with proprietary hardware vendors pushing MAID now.
I am sure there is a counter argument from MAID vendors for each of these points - perhaps people in the mainstream are just not convinced they have all been addressed, yet.
Some Sun customers have been hacking together their own code to do it, at the Solaris level, instead of the ZFS level.
I have personally been using external 1.5 TB drives on a SPARC server, which spin down automatically, under ZFS with RAID. I am personally very happy with the results (far superior to when I tried to run these drives under MacOSX or Windows.) I have been considering expanding the usage after designing/implementing a better power supply which combines batteries to eliminate multiple transformers plugged into a UPS.
I love the concept of MAID, I think MAID may be just too immature right now, or just too little too late.
The big MAID usage is yet to come
Some of the really cool potential applications for MAID has never managed to take off:
1. Online video media. MAID is the only cost effective way to manage the long tail for providers who do not own CDN infrastructure and use centralised delivery instead.
2. Off-site network backups. Classic MAID application.
3. Virtualised desktops and mandatory network storage (webtops). If you maintain desktops for several thousand people on a centralised array MAID-ing is an extremely good way to save energy without sacrificing performance.
All of these applications are in the class of "will happen any time now". However that any time has not happened for 5+ years straight and one of the reasons is surprise, surprise - the fact that CIOs continue to budget for RAID (or tape in the backup case) instead of MAID for all of these.
Another factor is the lack of bandwidth. If every SMB had 50M up/50Mb down a MAID based network backup service would have been commercially viable. Same for download video services, same for remote desktops for telecommuters, etc. Unfortunately the bandwidth is simply not there :(
I love acronyms
Bugger the MAID and themanfromars, where's the Total Wide-Application Terabyte Disc Array Nano-Gold Layer Electronics?
Three reasons why MAID can't fly...
Consider: it takes 7 watts to keep a 3.5" SATA HDD spinning in the "MAID" context.
It only takes 0.6 watts to keep a laptop HDD spinning. Laptop HDDs use 1/5th of the power-per-spinning-GByte than "MAID" disk drives, because power suck goes up exponentially with (a) platter diameter and (b) RPM. This is fundamental physics and is true forever, at least until somebody learns how to spin HDDs in a vacuum, which has been shown to be impossible (practically and commercially speaking).
Consider also...the "sleep mode" of a spun-down 3.5" MAID disk consumes MORE power than a laptop HDD requires when it is spinning (in "idle" mode)! Read that again, and if you don't believe it, go compare spec sheets.
Why on earth would people pay money for the privelege of "spinning down" MAID disks when they can simply use disks that don't suck power in the first place? Now, this begs the question "then where are the laptop-disk based D2D backup solutions?" The answer is...
Look at the ROI. Better yet, look FOR the ROI. You won't find it. That's because, based on the power savings, a MAID will pay for itself in about 43 years. Instead of asking "where's the love", ask "Where's the beef". You can scour Copan's entire site for a ROI analysis and you won't find one. The only place you'll see it in in the unsubstantiated claims made in their press releases.
Reason #3: DeDuplication
Deduplication kills the value proposition for MAID two ways -- it (a) reduces the raw storage requirements by 10x - 40x, yielding FAR better energy savings than MAID, and (b) it makes the remaining, de-duplicated data store that much "hotter", because there no longer are dozens or hundreds of duplicate objects out there as good candidates for an HDD spin-down. This wipes-out whatever benefits MAID might have offered prior to de-duplication.
- Hi-torque tank engines: EXTREME car hacking with The Register
- Review What's MISSING on Amazon Fire Phone... and why it WON'T set the world alight
- Product round-up Trousers down for six of the best affordable Androids
- Antique Code Show World of Warcraft then and now: From Orcs and Humans to Warlords of Draenor
- Why did it take antivirus giants YEARS to drill into super-scary Regin? Symantec responds...