Slashdot Mirror
Rethinking Computer Design For an Optical World
holy_calamity writes "Technology Review looks at how some traditions of computer architecture are up for grabs with the arrival of optical interconnects like Intel's 50Gbps link unveiled last week. The extra speed makes it possible to consider moving a server's RAM a few feet from its CPUs to aid cooling and moving memory and computational power to peripherals like laptop docks and monitors."
For GPUs? Finally an easy upgrade path for all future Macs?
moving memory and computational power to peripherals like laptop docks and monitors.
I would think that this would make upgrading more complicated, not less so. Thoughts?
Living With a Nerd
that should help reduce latencies
The extra speed makes it possible to consider moving a server's RAM a few feet from its CPUs to aid cooling and moving memory and computational power to peripherals like laptop docks and monitors
Why would I want to pay for computational power in my monitor? When I buy a monitor I want it to do it's job - show the best quality images for the cheapest cost possible. A good monitor should last much longer than the associated computer driving it (unless we suddenly have a huge increase in the rate of development of display technology). Why would I want added cost in my monitor that will only make it out of date more quickly?
moving memory and computational power to peripherals like ... monitors.
They mean ever more complicated DRM. Like sending the raw stream to the monitor to be decoded there.
"Science flies us to the moon. Religion flies us into buildings." - Victor Stenger
I mean for laptops. Right now I can leave storage and a larger monitor when I take it with me, and of course anything that can be networked. I'd like to be able to "dock the laptop into" more RAM, a more powerful GPU, and (while I realize this is wholly unlikely) maybe a second CPU (4 cores on the laptop, 4 more on the table).
Adding a GPU as an external peripheral has already been done, just not in a commercially viable way. Hopefully this will change.
Entomologically speaking, the spider is not a bug, it's a feature.
Same goes for optical interconnect to memory: the flood may be Biblical when it arrives, but while waiting for it to arrive the processor isn't doing anything useful.
Now, peripherals are another matter. But if bandwidth were all it took, we'd be using 10 Gb/s PCI Express for memory right now.
Lacking <sarcasm> tags,
The question is how many years it'll take before Windows supports it.
Without factoring in speed of light drops due to index of refraction changes, at a distance of 1 meter you're looking at latencies of 7 nanoseconds just for travel time. The bandwidth may be decent but the latency is going to be an issue for any significant distance.
They want their rats nest of cables back.
The extra speed makes it possible to consider moving a server's RAM a few feet from its CPUs to aid cooling and moving memory and computational power to peripherals like laptop docks and monitors.
Have gnu, will travel.
because this appears to be another aspect of Wheel of Reincarnation.
I'm old enough to remember a time where a computer was a series of bitty boxes tied together with cables. Then someone decided to integrate a lot of the stuff onto a motherboard, with just loosely-related stuff connected by cables to the motherboard. Then the loosely-related stuff got put into cards that plugged into the motherboard. Then that stuff just got integrated into the motherboard.
And now it's being reborn as stuff in bitty boxes connected together with cables.
I wonder what enlightement will be like, because karma appears to have been a bitch.
Welcome to the Panopticon. Used to be a prison, now it's your home.
The extra speed makes it possible to consider moving a server's RAM a few feet from its CPUs
Sure it has the bandwidth, but have you tried calculating the speed of light into that? Long ago I saw part of an interview with Grace Hopper, and she held up a six-inch piece of wire. She explained that the piece of wire represented a nanosecond delay. Now admittedly electricity usually only travels at about 0.5c, IIRC, but I think she was giving the speed-of-light delay, not the speed-of-electrons delay. I'm also not including any propagation delays in the optic transmitter and reciever. Also, the delays are doubled because the CPU has to request what data needs to be sent, and that has to arrive at the memory before the memory can send the data.
"A few feet"? Let's say 3 feet. That means 3 feet times 2 directions times 2 nanoseconds per foot, for a total of 12 nanoseconds, maybe a little better if you can make page requests. I remember back in the early '90s, RAM speeds were in the range of 60-80ns for plain old fast-page DRAM.
You can deal with relativistic propagation delays for secondary storage, but not for primary storage.
#naabhaprzrag, #sverubfr-000, #agi-fcbafberq, negvpyr[pynff*=' negvpyr-ary-'] { qvfcynl: abar !vzcbegnag; }
IANAEE (I Am Not An Electrical Engineer) Pardon my possible stupidity, but what was keeping us from putting the RAM a few feet from the CPU? The way I understand it, electrons don't move much slower than light. Of course you might lose current.
Yes and no. In a vacuum, electrons aren't terribly useful unless you're driving them with a particle accelerator. In wires, electrons aren't really doing the work anyway: electrical signals effectively travel as waves in the dielectric surrounding the wires and in particular between signal pairs. In that case, the signal travels at around half the speed of light in a vacuum (faster if you use expensive insulation like Teflon, slower for other plastics.)
Light in optical fiber is also slowed by the refraction coefficient of the material and by path-length extension in multimode fiber. However, on balance it's a bit faster.
The real gotcha is that electrical signals at outrageous bandwidths suffer from some really horrible losses due to both skin effects on the wires and dielectric losses in the insulation. At 50 Gb/s and 30 cm, you're doing well to detect the resulting signal, never mind decode it. Worse, the losses are highly frequency-dependent, so you have to do all sorts of ugly things to pre- and post-condition the signal to make it usable. Some of this can be overcome by cranking up the transmit power, but then you get into that property of wires known as "antenna." All of that processing at both ends takes time, too.
Just not worth doing, generally.
Likewise, putting a bunch of streams out in parallel requires all sorts of cleverness to put the separate lanes together again after transmission skew. A single optical stream is much easier to use, sort of like the communications equivalent of Amdahl's Law.
Lacking <sarcasm> tags,
My dream computer has always been a completely modular system, with every component accessible and hot-swappable. I always imagined it being about the size and shape as a normal computer, but covered in slots, with video cards, RAM, drives, etc in the form of cartridges... pin lengths designed to make sure the right things contact in the right order...
While lamenting the poor graphical performance of my laptop, I investigated external graphics cards. While they aren't currently suitable for... well... anything, a nice 50gbps optical cable might make it a plausible scenario.
I would even prefer an external video card for my desktop computer (if performance matched the internal version). It could have its own case, cooling, and powerbrick, instead of murdering my internal power supply, heating my computer up, screaming like a jet engine, and possible bursting into flames when my haphazard system design blocks vital airflow.
Bigger computers!
What we've been working toward all these decades!
1. The Internet already does that. How much of the experience today is processed partly in a faraway datacenter? I know that even users like me use the Internet as a method to pull things away from each other so each part lives where it makes sense. I have a powerful desktop at home that I RDP into from whatever portable device I happen to be toting. I don't worry about my laptop getting stolen, the experience is pretty fast (faster than a netbook's local CPU, for sure), and I get to mix-and-match my portable hardware.
2. This is going to have much more use at a datacenter than it will in a server closet or a home. I can already fit more RAM, CPU, and Storage than I need in a typical desktop. Most small businesses run fine on one or two servers. Datacenters, on the other hand, could really take advantage to commoditizing RAM and CPU, like they have with SANs in storage. No more 'host box/VM', it's time to take the next step and pool RAM and CPUs, and provision them to VMs through some sort of software/hardware control fabric. I think Cisco already knows this, which is why they're moving to building servers.
Imagine the datacenter of the future:
Instead of discrete PC servers with multiple VM guests each and CAT-6 LAN plugs, you have a pool of RAM, a pool of storage, and a pool of CPUs controlled by some sort of control interface. Instead of plugging the NIC on the back of it into your network equipment, the control interface is -built into- the network core, wired right into the backplane of your LAN. Extra CPU power that's not actually being used will be put to work by the control fabric compressing and deduplicating stuff in storage and RAM. The control interface will 'learn' that some types of data are better served off of the faster set of drives, or in unused RAM allocated as storage. 'Cold' data would slowly migrate to cheap, redundant arrays.
Guest systems will change, too. No longer will VMs do their own disk caching. It makes sense for a regular server to put all its own RAM to use, but on a system like this, it makes sense to let the 'host fabric' handle the intelligent stuff. Guest operating systems will likely evolve to speak directly to the 'host' VFS to avoid I/O penalties, and to communicate needs for more or less resources (why should a VM that never uses more than 1GB RAM and averages two threads always be allocated 4GB and eight threads?).
"Sometimes, I think Trent just needs a cup of hot chocolate and a blankie." -Tori Amos on Nine Inch Nails
In 30 years I'll suggest integrated optical motherboards.
http://arstechnica.com/old/content/2006/01/5971.ars
Potential applications: Subspace radio, wide area networks on a solar system scale. Just think, no more 3 minute wait for a radio signal from Mars or beyond.
Well, I'm not aware of anyone using epoxy glass for cable insulation. You can get pretty quick (0.8 C0 or so) with foamed Teflon insulation, but you have to be seriously wanting to pay for it. Easy to damage, too.
Lacking <sarcasm> tags,
Adding a second CPU is not that unlikely - motherboards with two sockets exist for a long time. If you can "push out" the RAM with this tech, why not a second CPU?
Dilbert RSS feed
Powered Partridge?
How much is your data worth? Back it up now.
That's what I was thinking: This is going back to the way it was in the mini-computer era. CPU in one box. Additional memory in another. Framebuffer in a third. Disk in a fourth...
What's old is new again.
Without disclosing my Super Secret Identity, let's just say that I was there at the beginning of the FBDIMM fiasco, told my management to run, not walk, away from getting sucked into it, and proceeded to watch the train wreck from very close up. As in, on the field instead of front-row in the stands.
I've made a lot of bad calls in my life but I totally nailed that one.
Lacking <sarcasm> tags,
For what it's worth:
c / (3 ghz) ~= 9.993 cm
Perhaps half of this is really the characteristic length (two way communication and all). I don't really know how RAM works, so with DDR it may even be half of that length, which puts it at about 2.5cm / 1 in. (roughly). I leave it for someone else to tell me why these numbers mean absolutely nothing (seriously, I'm not too proud to learn something here).
Uh yeah, this isn't the first time around. The computer industry is constantly rediscovering previous designs. Timesharing, batch jobs, client-server, intergrated/distributed processing, etc, etc. Nothing new under the sun, just smaller and faster is all.
I wonder what enlightement will be like, because karma appears to have been a bitch.
It's called retirement - you get out of the loop and eventually you go out like a the flame of a candle.
When information is power, privacy is freedom.
The vast majority of computers (even if known by other names such as "smartphone") will only become more and more integrated. I doubt we'll be buying standalone graphics cards for PC's in 10 years, and not even standalone RAM modules in many cases.
Maybe for high performance computing there will be a big shared memory hooked up to tens of thousands of cores by optical interconnects, but not for 99% of the market.
Except the whole thing has terrible power consumption, because each unit has its own crappy wall-wart power supply, and you have to have 3 power strips wired in series to have places for them all to plug in.
Wasn't Infiniband, 3GIO (now PCIe) and a plethora of other forward looking interconnects supposed to have already done this by the early 2000s? There was even talk of extending Hyperlink to a 10m range at one point.
Wake me up when it hits silicon.
so you can have lot's boxes that need there own power cord / big black box.
I don't see a data cable having the power to drive a good GPU.
No. Power is easy. Again, like the old days... One large power supply module with a cable jumpering modules together. Then it can be as efficient as you want.
50 Gbps is bandwidth. What's the latency? That'd be kinda important for the purposes of remoting RAM.
You could have one shared PSU. Nobody says that the cable connecting the individual boxes must be optical-only; you could have the PSU transform it down to 48 V DC, doing DC-to-DC conversion down to 5 V/12 V in each box. This could let you make the power cords for each box thinner, allowing you to use a combined data/power cable and connector. Then just spec out the power distribution such that daisy-chained devices will be connected to the PSU in parallel so that one breaking device won't take down everything else; also, make hotplugging mandatory.
Voila, an easily extendable multi-box computer using only one additional wire per additional box. As an added bonus you could completely destroy your GPU and (assuming the OS supports it) the rest of the computer will happily hum along in a headless configuration until you've replaced the GPU box.
USE HOT GRITS WITH STATUE OF NATALIE PORTMAN (NAKED AND PETRIFIED)
Won't work: you'd have to get all the modules from the same manufacturer. That's the only reason it worked way back then: in those times, all the parts came from one company (Commodore, Apple, etc.). There wasn't much compatibility in computer parts in those days.
It won't work today because we don't have vertically-integrated manufacturers any more (except Apple, and they're a niche player). Doing such a thing would require a standard. That's about as likely as the different manufacturers getting together and deciding on a single standard for the higher-capacity replacement for the 1.44MB 3.5" floppy. That never happened, and floppies finally were obsoleted by CD-Rs, networks, and finally USB thumb drives.
Why don't we just go back to the "old" method of having one big box (like a "tower" style) that has everything in it, except for the keyboard, mouse, and monitor? It makes it a lot easier when you have to move your computer, and keeps your desk from being a giant jumble of cables.
I'm not a computer engineer, so I don't know whether this is the step we need, but I've always wanted a laptop and desktop that would dock together and increase their speed when they were connected. Does that make sense? Obviously, syncing data and setting between them so that they could operate separately would be quite a chore, but I think it would be cool. Two slow computers, with one being portable, or one fast computer when you're at your desk (could we even have multiple desk units and sync to the portable?). Fun stuff.
The big benefit is the few small optics for photonics over the many small wires for electronics. Reduces interconnect space and the number of things that have to succeed in order to have something sellable.
http://www.google.com/search?q=c%2F1ft
It takes light about the same amount of time to travel 1 foot as it takes a 1 GHz CPU to complete one cycle. This places a hard limit on how far apart your components can be even with zero switching/decoding time if you're running on a modern multi-gigahertz server CPU. Thus, the travel time is actually extremely important and as the article notes (and the summary is wrong about), this technology will allow the memory to be moved maybe a foot away.
Based on how people adopt new technology, my guess is that people will just try to pack more memory onto the same board rather than improve the ventilation.
But remember this tech does not change the speed of light, every foot is still approximately 1 nanosecond (every 30.5 centimeters approximately for the non-USA parts of the world). So each foot adds 2 nanoseconds (As JRRT said, There and back again) which through clever local cache might get down to much less. But the latency is still there regardless of the block size transferred.
- Tjp
I am in wallow with my inner money grubbing capitalistic pig. ... Oink!
RAM speed is *far* more sensitive to latency and transaction times. I'm betting that to get 50GBps you're seeing lots of latency (as compared to traditional RAM interconnects - remember we moved those from south bridges, to the CPU, just to shave off that bit of latency in recent gen CPUs!)
Sure it will. Look inside that big box. Lots of components made by various manufacturers. One power supply built to a standard. Standard voltages = 12V, 5V, 3.3V. Everything uses one of the above and has a standard connector. Any other requirements are created by the component as needed, from one of the standard voltages.
As to the "old" method of having one big box... You don't go back far enough. Before the "old" big box, there were older stacks and racks of separate boxes.
To be honest, I suspect, for the most part, it will remain in one box. Intels proposal is for a modular approach that could very easily be contained in a single system chassis.
Sure it will. Look inside that big box. Lots of components made by various manufacturers. One power supply built to a standard. Standard voltages = 12V, 5V, 3.3V. Everything uses one of the above and has a standard connector.
That's mainly because the power supply standards were designed by IBM with its PC (such as the 4-pin Molex connector on drives), and then later the ATX standard driven by Intel, which retained the Molex and the 3.5" floppy power connector, and added a new motherboard power connector, and later a 12V CPU power connector and a SATA power connector. Standards have to have some big company forcing everyone else to use them, or they won't catch on.
So this might work, as long as Intel specifies the power connectors as part of the standard.
But you can't rely on a bunch of competing companies to create a standard. They tried that with the floppy replacement, and it was a disaster. IBM tried to push the 2.88MB floppy (with their PS/2 computers), but by then IBM was a has-been and were mostly ignored, and no one cared about IBM compatibility any more (e.g., the MCA bus was mostly ignored). If Intel had pushed a new standard, that might have caught on, but Intel didn't care much about removable media.
Just look at how, for many standalone components like cable modems, routers, HDD docks, etc., they all have completely different power connectors and power requirements. If you lose the wall-wart or it dies, you might be able to find a compatible replacement, but it won't be trivial and you certainly can't use one of your other ones. And there's no way to share a single wall-wart among several devices, even though this would be more efficient.
the IBM 2.88MB floppy came out at the same time as the Iomega cartridge drives and a number of floppy sized optical drives. There were a number of better solutions; none of which were compatible. Also available at the time were inexpensive write once CD's. None of these were even remotely compatible, which is why there was no consensus. Personally, I used the CD's (I still have a spindle each of blank 80cm and 120cm CD-R disks).
I usually don't have a problem replacing wall warts... but I'll just jump on Digikey, order the adapter with the right output and a connector. I also save all the old ones and re-purpose them as needed (again, replacing the connector if necessary)
the IBM 2.88MB floppy came out at the same time as the Iomega cartridge drives and a number of floppy sized optical drives. There were a number of better solutions; none of which were compatible. Also available at the time were inexpensive write once CD's. None of these were even remotely compatible, which is why there was no consensus. Personally, I used the CD's (I still have a spindle each of blank 80cm and 120cm CD-R disks).
I seem to remember the IBM 2.88 coming out a little before the others (esp. the 21 MB "floptical"), but mainly being ignored though many machines supported them in BIOS. I also seem to remember the CD-Rs coming out a little later. The CD-R came out quite a bit later, and when that happened, that pretty much killed off the Iomega Zip and LS-120 and everything else.
According to Wikipedia (http://en.wikipedia.org/wiki/Floppy_disk), IBM's 2.88MB disk came out way back in 1987, the Floptical in 1991, Iomega's infamous Zip drive in 1994, and the LS-120 in 1996. As for CD-Rs and CD-RWs, the latter didn't come out until 1997, and while CD-Rs were available as early as 1990, a system to use one cost $35k, and CD-R drives didn't fall to under $1k until 1995. It wasn't until the late 1990s that the drives and media became so cheap that people abandoned floppies for them.