Slashdot Mirror
16,000 CWRU Computers Getting Gigabit Ethernet
lowlypeon writes "In a move that makes going back to college more tempting than usual, Case Western is installing fiber connections in 16,000 computers over the next year to give students a 1 gigabit per second Ethernet connection. Administrators aren't sure what anybody needs that kind of bandwidth for yet, but they are curious to see how it gets used."
This is intereresting but not that all unexepected --- Case has always been on the cutting edge of networking technology. I almost ended up going to school there just because they already have ATM Fiber conncetions in all the rooms already (155Mb/sec IIRC). Gigabit ethernet in itself is interesting because I Don't know of any hard drives that can actually move 100 megabytes a second conitinuously, but I'm sure it will catch up one day. I wouldn't be surprised if students started building RAID striping arrays just to get the bandwidth up.
On another note, the article doesn't seem to mention the speed of the actual internet hookup. Anyone know what they have over there? If they haven't upgraded that, then the whole thing won't seem any faster than 10-base-T when using the net.
"Western Reserve" is a term that was left over from the old original 13 colonies. When the original colonies of the United States were formed, most of the western borders were left blank, since the settlers didn't know how far west the land went. In 1786, Ohio was actually part of State of Connecticut. Connecticut gave up its claims to Western lands of the United States, except for a portion of northeastern Ohio known as the Connecticut Western Reserve. Later, the land was sold to the Connecticut Land Company, which surveyed and settled the region, but the name Connecticut Western Reserve - or just Western Reserve - continued to be used to describe the northeastern section of Ohio.
--
Does anyone remember
"Administrators aren't sure what anybody needs that kind of bandwidth for yet, but they are curious to see how it gets used"
I am interested to find out how the administrators will find out how the links are being used. What hardware/software has the power to track this level of traffic on a switched network.
Will they be using integrated RMON2 NAM cards in the switches or possibly analysis of Netflow data from the routers.
However they do it there will be a geat deal of data to crunch
slashnik
A 3c996 SX card runs about $475 at discount
:)
Seriously, who is selling these people 16,000 fiber NICs that they will need to hook up all the PCs? Or are the students expect to foot the bill, on TOP of the $400 per year tech fee listed in the article?
Wouldn't this make a lot more sense?
Newegg.com sells retail boxed, Intel Gigabit cards for $55. So the question is now, how much is the fiber to giga-copper transceiver?
SlashSigTheorem: Humorous, Political, Critical, Constructive- If you have a
Where I work, we just got around to upgrading the network from 10BaseT to 100BaseT. Things did get faster -- but not ten times as fast. In a few extreme cases, apps that transfer umpteen megabytes in a short span were maybe 2 or 3 times as fast. Fifty percent was more typical.
Case was once famous for doing the wrong thing really well. Just as interactive computing started to work, Case developed one of the best batch operating systems of the era. This was so cost-effective that it kept Case studends on punched cards much later than other comparable schools. The entire school ran on a 1 MIPS machine, with enough free time to support a private company selling excess time to commercial users.
Case Western used to have an ATM network available to all of its students which was all super fast and had high bandwidth. However, they found that a pure network of ATM was too unstable (there were numerous random outages that got really annoying) and started switching back to ethernet.
However it looks like they are not going to settle with 100mbps and just go straight to Gigabit.
I guess they like to have their network be as fast as possible.
As another alum to CWRU, all I can say is I hope they do this upgrade right this time. I was taking classes there during the "upgrade" to ATM and while everything was far faster than the typical 28.8K connection of the day, there were alot of bandwidth problems due to their approach to implementing the ATM. Although I only used the ethernet connection, and not the ATM, (First PC was only a 486, the second was a PII 233, but at the time they didn't have reliable ATM drivers for the PII architecture) and the bandwidth for ethernet users was always much higher than your typical ATM user. Why? The infrastructure of the CWRUnet ethernet set-up was well established and optimized from the central core servers out to the routers for each building and dorm. The huge mistake of the ATM system was they tried to set it up all at once, instead of testing the stability of a core set-up before branching out to every other computer. The result was an absolute mess as far as lost email and dropped packets left and right. All of this seriously effected everyone from the astronomy dept generating 1 million point galaxy models to the untold hundreds of Quake and StarCraft games going on over campus. So, now that they have a new guy at ITS running the show, lets hope they do better this time.
FUD! FUD! FUD!
:)
As a student worker who has helped trace some of the fiber on campus and done work on several of the network maps, I believe I'm qualified to answer this to some extent.
The Mail server - yes, it is still unencrypted, although that should end with IMAP this fall. (in theory). Besides - the gigabit is all switched. Much harder to see what's going down the wire unless you MAC spoof.
As for the card access - i'm not sure about the past, but currently the card readers are on their OWN SET OF FIBER. The SSNs are sent as SERIAL data. The card readers are serial devices that come back to FOMs (fiber-optic modems) which run singlemode back to Crawford at the end of the quad. From crawford, all of the signals are muxed again onto one pair of singlemode that is then shot over to security across Adelbert, which houses the server that actually holds the access records for the buildings.
As far as I know, card access has always been done like this. there has been talk of running it over IP, encapsulated & encrypted to hell. However, that is just talk, and it hasn't been done, yet or ever. Your SSN isn't readable. Anyone who says otherwise has either tapped the actual fiber that runs the cardreaders, or the layout has changed from the past.
From a motherboard manual, error beep codes: S-L-L-L-SS: Speaker Error
1 Gbps is comparable in speed to the internal system buses -- it's 125 MegaBYTES per second!
In effect, you can now think of all the computing resources on the LAN (CPU, disks, memory, etc.) as being interconnected with a fast system bus comparable in speed to PCI. 10 PCs connected with GigE don't look much different than a multiprocessor with 10 CPUs, do they? And that's with off-the-shelf PCs that were built with today's assumptions that network bandwidth is limited. As bandwidth becomes more plentiful, those assumptions will change, and PCs will add hardware and software support for all sorts of cool stuff...
There are still a lot of unsolved problems in this area, of course, which is why it's so cool to throw these problems at a university full of smart students.
Some that I can think of from the top of my mind are:
o storage connectivity: in a traditional PC, your hard drive tends to be local, because you need
several megaBYTEs/second read/write bandwidth. Well, with GigE, you now have enough bandwidth to write to a disk connected to the network. This stuff is still in its infancy.
Do a Google search for "iSCSI" -- SCSI over IP.
o multiprocessors: Someone mentioned Beowulf clusters, which I can see as a first step to getting some parallel computing going. But there are still bottlenecks, like the software overhead of transferring the data around, computing TCP checksums, etc. etc. Think how cool it would be if you could just access a remote machine's memory like you can local memory. You need special hardware support to take away some of the overhead. This stuff is also still in its infancy.
Do a Google search on "RDMA" (Remote DMA).
o Someone asked "why GigE to the edge?"
One reason why networking companies are pushing GigE all the way to the end hosts is so that the network elements (switches, routers, etc.) don't have to buffer bursts of traffic when you're doing speed conversion from a 1GigE uplink to a 10/100 Mbps end host. If a burst of packets arrives at the edge switch from the GigE uplink, they'll have to be buffered for some time while they "trickle" to the endstation. If the link to the end station is also 1 Gbps, then the packets can be transmitted as quickly as they arrive (assuming that the link is not oversubscribed).
Why do you want to avoid buffering in the network?
Buffers in switches and routers are MEGA-expensive, because they have to be REALLY fast. We're talking memory bandwidths of 8 Gigabytes/second on a non-blocking 32 Gigabit/second switch fabric. That stuff is really hard to build, and even harder to build cheap enough that anyone can afford it.
Buffering introduces delay (latency).
o Also, keep in mind that modern switches can apply QoS mechanisms like policing to limit the amount of bandwidth that a given host can consume, so the fact that your "pipe" can "do" 1Gbps doesn't mean that you'll be able to get that much traffic into/out of the network. Depending on the switch, administrators can configure how much bandwidth you get pretty finely.
Actually, gigabit ethernet has higher bandwidth than the PCI bus, which is amusing considering that we're getting gigabit PCI cards.