Commentards Ahoy! How about a Petabyte of storage? A little experiment from us: we are rounding up comments on a couple of articles - and turning them into articles. Some might call this stretching the material - social media types might call it amplification. But we think that not all of our commentards are mad - and many have very interesting insights to share. Today, we mine …
I can remember doing a program is machine language and it took 47K of space to distribute. Today the same program, with the same functionality, takes 2.8 MB PLUS the need to install .NET 3.5 (another 230 MB) for all the needed runtimes.
More space to counter programming inefficiency is inelegant at best. Users will consume whatever space you give them.
"Users will consume whatever space you give them."
THIS, So very very much. This is known as Parkinson's Law of Data
We gave our mail users two servers with 500 GB mail spool space each at the last upgrade- the old kit was a single server with 400 GB total.
I've been fighting a (losing) battle the past three weeks with the new servers running out of space. The new servers are maybe 18 months in service?
Unless you've got some kinky fetish I willing to bet you don't all put on stilettos and stand in the same square metre of space.
It's not just the room taking 400Kg, its the four square centimetres under the feet. But I get your point, some people do overplay the reinforced floor thing.
Paris: because she knows stilettos! And group hugs.
400kg, or 4kN approximately is a lot, depending on how it's sited. A residential floor needs to be able to support 1.5kN/m^2 (BS6399). Say it's all mounted in a single 19-inch, about 0.6m x 0.6m means it's generating a load of 11kN/m^2, which is substantially over that limit.
I don't worry about entering a room with 4 people in there, but equally I don't stand in the same area as all 4 of them. As long as the room is over 2.7m^2, then you'll be fine. And fitting 5 people in a room that small isn't that common.
Not to say it wouldn't be a problem, but certainly don't bank on it.
Everybody is obsessed with the floor loading because we have heard horror stories from each other about hundreds of thousands of pounds worth of equipment falling through a floor and landing in the room below somewhat worse for wear.
That, and most people employed in IT are paranoid (it's a positive career trait in IT!) Instead of thinking, hey, what are the chances, why worry about it? we tend to think more along the lines of what are the chances? being the first line of a risk assessment, even if that is just a back of the envelope exercise concluding that we are more worried about a meteorite taking out the office.
I did like the comment about offline storage. I actually considered a disk library project a while ago along similar lines.
My plan was to pack 4 3.5" disks into a "module". This module would be a RAID-5 set, but would be used like a tape in a tape library. Files would be written to it, and their location recorded in an online database (as in on a hard drive in the machine, not as in a server on the internet). Need a file, it's looked up in the DB, you are prompted for the module number. Insert it and you've got your data.
Using today's prices (quickly gleaned from ebuyer), the storage itself using 4x3TB / module @ £100ea would give, in RAID-5, 9TB for £400. For 1PB, you would need 112 modules, leading to approx 45k. Add in a grand (over estimating) for a server to deal with it and you have your PB of storage, offline but redundant.
A caption sketch in a 1980 copy of Personal Computer World:
geek1 "so, they've produced a chip with 10,000 transistors on it"
geek2 "yes, now they just have to figure out how to connect them together!"
Give it 15 years max, and your colleagues will gaze at you with pity as you fondle your 1Pb key fob
This really does need to be local storage - forgetting security issues for now, the cloud really isn't an option for most of us for this. My broadband (BT finest 8Mb download) uploads at about 400kb/sec, which is about 6GB/day or 2TB/year - so 500 years to upload a petabyte! Who knows, we may have FTTC by then...
The current high density enclosure is 5U84 so with 4TB HDDs it gives 336TB.
3 of these will get you 1PB but let's add a 4th so you can run RAID 6 and have a few hot spares (experiment with HPC file systems maybe?)
That gives you 20U, about 600kg and can just about run from a 32A 415V 3-phase supply. It also kicks out between 6kW and 10kW of heat so turn the thermostat down and you may want to move a few houses down the street because cooling this lot is going to be loud!
Something wrong with that heat calculation.
1Pb = 360 (say) 3Tb drives. Power consumption of WD 3Tb "Red" active read/write = 4.4W. http://www.storagereview.com/western_digital_red_nas_hard_drive_review_wd30efrx . 360 x 4.4 = 1584W. You can run that off a single 13A domestic outlet, even allowing for PSU inefficiency and all the cooling fans.
You WILL have to make sure your domestic Petabyte server does a phased disk spin-up, because the power demand of a disk drive while it's spinning up is usually about 20W for about ten seconds. 360 x 20W is 7.2kW, about the same as a somewhat weak electric shower unit, so you'd just have to spread the load over two domestic circuits and four 13A plugs if you weren't sure it would never spin up all the drives at once.
BTW those WD Red drives are extraordinarily quiet, so you wouldn't need ear-plugs provided you could find some enclosures that used big 12cm fans, rather than huge numbers of 6cm fans.
"360 x 20W is 7.2kW, about the same as a somewhat weak electric shower unit, so you'd just have to spread the load over two domestic circuits and four 13A plugs if you weren't sure it would never spin up all the drives at once."
You wouldn't need two circuits; a standard "ring final" in the UK is fused at 32A and 7.2kW is 31A @ 230V. You'd probably need to use three plugs (each fused at 13A) to get the power to the racks, though it would be interesting to try pulling 15A through a 13A BS1363 fuse. I'm fairly certain that it will get warm but not fail, at least, not for a long time.
Of course that's assuming that your 20W at startup is correct - isn't it much more likely to be some kind of peaking curve? Might need to swap your B-curve MCB for a C-curve MCB if the transient is large for any appreciable length of time :-)
For those countries not blessed with the UK's rather "interesting" way of wiring houses I believe standard sockets radials are either 16A or 20A (you can do those in the UK too) so, yes, in those circumstances you would need two at 230V.
110V? Don't even go there.
you can buy a 4TB external USB3.0 drive for $210.00 USD. A 10-port hub costs $50 USD. A computer with USB 3.0 and a 10 Gig-e NIC costs $1200 USD. An 8-port 10Gig-e switch costs $1000.
one switch, 7 computers, 25 USB hubs, 250 drives: $1000+7x$1200+25x$50+210x$250=$63,150.
Now double it because we want a second one in a secure location for backup.
The external disks are 2" wide and < 5" deep, so 40 sit on a 4' shelf 1'deep, and we need 7 such shelves, about 8" high, and each has room for one computer and four hubs.
For power, we need a cheap way to turn the AC power on and off for the disks. Unfortunately, its not cheap so we turn them on and off in sets of 10. To access the data, turn on the correct set, copy the data, turn off the set. power control for 25 sets will cost perhaps another $2000.
Update on power control: Its ugly, but a single-circuit USB-controlled unit costs $25 USD. So, 25 x $25 costs $625 USD, not the $2000 mentioned previously.
The costs mentioned in the original post were retail qty 1 on the web. I suspect you can get at least a 15% discount for this large order, so the total for each of the redundant 1PB systems is just over $50,000 USD.
Also, the architecture as stated is 7 systems, one per shelf, each capable of supporting 40 x 4TB, but actually supporting 36 to get an even balance. It might be prettier to use 8 shelves each supporting 32 x 4TB, just because it's binary. Adds slightly to the cost, but those $1200 computers are WAY overkill. In fact, since we are only powering up one set ot 10 at a time, we can actually get away with a single computer instead of 7 or 8 computers, which means we do not need the switch.
In case it's not obvious, this system is basically an automated version of a stack of unpowered disks on a shelf. Access time to a file will usually be about 10 seconds to allow the disks to spin up and stabilize.
Yes. Each of the 4' shelves needs four disk power distribution systems (DPDS.) Each DPDS would be a 10-position power strip plugged into a USB-controlled plug. The four DPDSs plus the four USB hubs plus the computer plug into a 10-position power strip, so each shelf ends up with a single plug leading out.
The main bulk is in the DPDS power strips and their plugs. Since each disk has a power cord that unplugs from the disk unit, it's possible to build a custom DPDS by cutting the plugs off of these cords and screwing the wires directly to a terminal block inside an approved small electrical box. All of this fits between the two rows on disks on the1' wide shelf.
I've got a grand total of about 100TB of uncompressed audio, photo, and video squirreled away on optical and rust based media. That'll do for now. If I was plastering the walls with 4K video, the current bulk price of 2TB hard drives compares quite favourably to the cost of a gloriously unreliable 1gb Jaz cartridge back in the steam age. I'm all about the incrementalism, daddy-O.. Ii was a bit of an effort ripping, burning and storing that lot. I needed a few thousand drinks to calm down.
We'll probably all have a petabyte of data subdermally implanted in our palms in 20 years. I'll wait till then to see how many 2PB nanoSD cards I can fit in a snuffbox.
Hmm,. I have my doubts about that 1.5KN/m^2 thing....
80 kg person (0.8kN), assume foot area of 30cmx15cmx2 (2 feet), gives 0.09m^2 and a total force (just standing still) of 8.888kN/m^2. Should I be worried about falling through the floor? Standing on one foot would be right out!
Assuming that averaging over the room is OK, then you have no problem with your 400kg unit, even with 5 people also in the room, as long as the total size of the room is 5.4m^2 or more. Or a room 2.3m on a side, which is not exactly large...
--Rakhal
I missed out the other condition as it was less relevant, but it also has to support 1.4kN point weight, so you'll be fine standing on one foot ;-)
And yes, as pointed out already 400kg would be fine in a room, but only if it were spread out over the whole floor, which it probably wouldn't be. And it won't necessarily happen the instant you subject it to that weight.
Source, BS 6399:
http://totalglassfittings.co.uk/upload/file/bs_6399_part_1_loadings_for_buildings.pdf
You'll see in that doc that server rooms have to be rated to 3.5kN/m^s and 4.5kN point weight - about 3x the strength of a domestic floor.
A petabyte of data has got to be predominatly "media" (audio/video) because there's no way average joe takes up that much space with, for example, emails, or source code or holiday snaps, or anything other than "media". So lets call it like it is: 0.95 petabytes of porn and 5 terabytes consisting of every thing they've ever written AND every picture or video they've ever taken AND every email they've either sent or received AND every program they've ever installed. And the thing about "media" is that it's essentially write-once - you're not going to be making changes to your "Terminator.avi", or "Billie Jean.mp3".
So my solution is to store your petabyte of data on tablets. Not the kind you buy from Dell or Apple. The kind make out of clay.
Build houses out of your wall-sized tablets. Arrange your kilobytes and megabytes into city blocks. Then offer them as free housing to the homeless, who then have a vested interest in making sure your data centre is secure. Sure, the read/write speeds would be terrible and it would be hideously expensive. On the plus side though, your data centre would end the world's homelessness problem. And you could use one house in that city to store a PC with a couple of 3TB HDDs to store all of the re-writeable/personal storage you'd actually use.
Mostly because they have no idea of the value of anything and when money and space are not an issue they wont even attempt to make sensible decisions on whether something is worth keeping or not.
The article (or peoples comments in it) make a really interesting point, that as file sizes grow we will naturally migrate to petabyte levels of personal storage - I have little problem with that ... you can't argue against trends and if the latest crap film takes 500gb+ rather than 4gb ... the only issue will be whether the storage media has come down in price enough to make physical ownership of that film a viable option. Personally I think this trend towards massive file size might well be the winning hand against the pirates, and we'll only have the option of streaming.
However, just imagine if file sizes stay roughly the same exactly how many files can be stored on a petabyte ... so many in fact that the ordinary person will have difficulty in finding what they want ... not such a problem if it's your own personal storage and you're really good at organising stuff ... but when you die and you will ... do you really expect someone to trawl through it all.
Lets face it they'll format it and and cart it off to their house. Just like the house clearance guys just throw everything into the dumpster so will all the digital tat that you've hoarded over your life go.
I know most people don't want to admit this but 99% of their digitally captured life is worthless boring shit, that no one not even their mum or kids are even remotely interested in viewing. Nothing has changed since the days of the photo album.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
