Slashdot Mirror
Interchangeable Data Storage Bricks?
shokk writes "EWeek is reporting that IBM is working on a concept called Ice Cube Storage Bricks that uses a conductive ceramic or mylar plate to transmit data between bricks across an air gap. Research center staff member Robert Gardner says that the idea is 'to walk up to the system, attach the storage and then walk away.' No mention is made of what happens when a brick in the middle of the cube needs to be replaced and the whole thing needs to be disassembled. To be really effective, this would need to be teamed up with some sort of a backplane, but the tech is new and neat."
Well, except for where it was specifically mentioned in the article.
It's getting bad when the person submitting the story doesn't even RTFA."Da ist ein Technölüst in mein Unterpanten!"
If you have to replace a block in the middle and the pile collapses, does the server crash and you lose?
Reviews with a twist! http://www.sardonicbastard.com
"EWeek is reporting that IBM is working on a concept called Ice Cube Storage Bricks that uses a conductive ceramic or mylar plate to transmit data between bricks across an air gap. "
Kind of like a neuron.
I knew that playing with legos would come in handy sooner or later.
I Am My Own Worst Enemy
How long before we see sites dedicated to storage array building contests?
That name for the individual bricks, coupled with the fact the picture they have on the website of the partially constructed collection looks kinda like this is rather disturbing.
I Am My Own Worst Enemy
the term "air borne viruses".
An Indian-American Hindu committed to non-violent thought/speech/action alarmed by the global explosion of radical Islam
Caveat: I'm an employee of IBM's Storage Systems/Technology group, but I'm not working on that particular project. I am only discussing things that were in the previous press releases about this product so you won't get anything confidential out of this post.
i c_storage/CIB_Hardware/
The original intent, when this was previewed a year or so ago, was that dead bricks would just stay in there and not require disassembly. See http://www.almaden.ibm.com/StorageSystems/autonom
for some more discussion.
The concern I have (my role in storage systems is error isolation and recovery) is that when you are running all these individual cubes, each one is trying to isolate what might have happened to its peers (or to itself) and when an error starts to propagate from one cube to the next, which it will invariably do sometime, you could end up with multiple cubes saying "IT'S THAT GUY!" and shooting him (ie, cutting him off) when in fact it was yet ANOTHER cube that started the whole thing by corrupting a message and is innocently sitting there not showing any failures.
So assuming that situation occurs, you have 1 failed and 1 not-failed cube which need to be fixed, and shutting off the failed one requires removal, which isn't part of the service model for the product. Needless to say, I'm going to be REALLY impressed when they get this working. My peers at IBM are awesome when it comes to storage, so I'm actually not being sarcastic when I say that.
...nother brick in the mass storage unit.
Darn. Doesn't scan.
You can already fit about 2TB is a large desktop case. These cubes only store 60GB/cube.
I would rather use loads of desktops, each with a local RAID array. Depending on bandwidth needs, I would either connect them to a common gigabit ethernet router (not so scalable) or set up dedicated routers in a tree heirarchy with larger and larger pipes as you get near the root.
Scalability should not be too much of an issue, and with 10 or so HDDs in a single case, you don't waste too much electricity.
Naturally, they would be running Linux.
Wait one damn minute here.
As they say in the army: "If it's stupid but it works, it's not stupid."
We spend billions on research, and only a fraction of the technologies that we invent (yes I am an IBM employee) turn into real products, but that's the whole idea.
Think of copper interconnects. Think of the 'pixie dust'. Think of the Power5 architecture. All of these things are working their way into YOUR badass PC of the future. These weren't the only things we came up with, but our process DID create them.
We must look really far forward and not sit on our laurels, that's a great way to lose the game against our competitors.
But as a people are mentioning, what about maintenance. You have a big stack of cubes, with something wrong in the middle, you have to dissassemble a bunch to get at it. And even if the data is mirrored on another brick, what happens when you have to remove that brick to get at the dead one.
Seems to me the most efficient manner is a two dimensional spread, i.e. cover one wall with cubes to a depth of only one cube. But then in that case, you migh as well go with a traditional rack server.
Cubes are stupid especially because (think about this) even if you have a wall of cubes stacked up, if you remove one, you may have to remove all of them on top of eachother (how are these things affixed?) Wouldn't it be smarter to have it be hexagonal? Removing one wouldn't collapse a column. There are more networking connections too.
Seems like a gimmick to me. Play tetris with your SAN!
That's a pretty trite answer but I'm on vacation, so I'll respond.
Let me throw it back at you this way: assume you are SPRINTING a marathon, one that lasts a whole year. You are contractually obligated to run as fast as you can, at peak speed from the start to the end and you can't stop for anything, you have to eat and drink and eliminate on the run.
Now let's say you catch a cold, or the flu in the middle of the race. Your biological system starts to steal resources to increase white blood cell counts, to fight the infection and eliminate it. In the meantime you are slowing down, staggering, unable to concentrate on the task. The flu can take DAYS to eliminate from your system. In that time, you have failed to fulfill the contract (ie, maintain max speed) and lost the race.
These are the conditions placed on the vendors of enterprise level storage systems. Modelling error recovery after biological systems doesn't work that easily. You MUST FIX THE ERROR RIGHT NOW when it happens and reject the failed part and isolate it from the system. We strive to do it in mere seconds, which corresponds to just shy of the timeout values of the host system that's trying to use your storage.
When you stub your finger, it doesn't fall off immediately because it MIGHT inhibit your ability to work at max efficiency. Yet, that's how it has to happen in storage. We see a part start to fail, and blammo, we shut it off and call for service.
They also talk about water cooling this system. Those connections are even harder to deal with. Hoses are always going to be thicker and more difficult to handle and there's the possibility of leaks, especially when connecting and disconnecting hoses.
I know that they are evil, because I've seen what they do.
...the answer is: It's all automatic. The bricks rearrange themselves in mid-air, and the broken bricks fall out.
Just watch Laputa. Near the end of the movie,- you see that Laputa is composed of these very same intelligent brick computers.
In answer to the question: "How do you replace the broken bricks in the middle?"
It's true.
Just watch the movie; It explains everything.
NASA spent untold dollars inventing a pen that would work in zero-G. The Russian's used a pencil.
Actually, I'll tell you how many dollars nasa spent developing a pen that worked in zero g. They spent 0. Not a cent. Someone developed it on their own for scuba diving, and then they bought tons of them from him because pencils (which they were using) have all sorts of problems in zero g - for instance, graphite and wood shavings could get in circutry, both are flammable, etc.
I submitted this story last night, and it didn't get posted.
Nowadays many data sources must run 24/7 because they serve data worldwide. It really shouldn't be necessary to shut down the data source to replace a disk. RAID on SCSI can already support hotswapping. The only nuance here is to make sure the physical design of the array allows the operator to replace any cube without removing others.