Cheap in terms of an individual funding it themselves, for research.
This seems like a nearly impossible task, given the size of current drives.
Cheap in terms of an individual funding it themselves, for research.
This seems like a nearly impossible task, given the size of current drives.
Although SvenW's is right in that you'd need ~340 disks (each drawing around 15 watts of power btw) and thus the raw disk cost, the hard/complex bit would be the controller and cabling management. Even if you used those 24/25 disk eSATA enclosures/extenders (which need power too), you'd still need 14 of those, most disk controllers have no more than 2 eSATA connections each so unless you wanted to daisy-chain them (bad idea) you'd need 7 of those cards (each drawing power too). So you'd end up with a machine with a lot of cards, some on wider buses than others, oceans of cabling and piles of power-hungry, hot individual disk boxes. And designing some form of RAID protection into the whole thing could get seriously complex, especially given you'd want to design the thing to survive a single disk, single enclosure and single controller failure.
I do big storage and as far as I'm aware big and cheap equals long term headaches.
edit - out of interest I ran some numbers;
Oh and I found a card that could in, theory anyway, connect to that many enclosures, not found out yet if it'd support that many disk though.
Oh and that controller can apparently deal with that many enclosures and disks - who knew! would be a lot of contention though.
Hope that helps.
Define "cheap". Backblaze have a famous blog post outlining their pod system for massive storage, but be aware that IO access will be slow.
You would need 16 backblaze pods to reach 1Pb of raw capacity, more like 20 or so to reach 1Pb usable capacity
Think about this: You will pay at the very least $20000 for the disks alone (without any redundancy, 340 x 3 TB x $60), without any cases, servers or controllers.
No problem if you want to experiment with a new way of storage. Look at Ceph which is designed for that.
If you now say "No, that's not what I'm looking for" then please specify your needs except "cheap".
The only way to do it right is in a SAN. Unfortunately they aren't what you call 'cheap'.
If the goal is 1PB of storage, then the above numbers give you a lower bound of something under $100k to something on the order of $1mil. Anything that involves 400 give or take spinning disks, controllers and cases to hold such disks, etc, will not be cheap either to run, cool, or maintain since there is close to a 100% chance than at least something will be broken today :-) This is unless you go with some solution designed to do 1PB of usable storage but requirement #1 for those sorts of solutions is "buy or lease some computer room."
Keep in mind as well, it takes just a little while to write 1PB of data. It takes on the order of 3 hrs to write 1TB of data, give or take, so about 125 days if you don't do it in parallel. Reading is on the order of just as fast. So, backup is a bit slow too.
If what you want to do is experiment with a distributed file system, then, do it with fewer and smaller disks. Pick up some oldish computers off of ebay, stuff some 1TB (or what ever is the cheapest $/GB disks) in them, and start researching.