The ECN No Name Newsletter is no longer being published. This is an archived issue.
[previous article] [next article]With the proliferation of workstations and servers on the Engineering Computer Network, users may have noticed that the network map is getting larger and more confusing. Along with problems understanding the network map, many users have noticed other, more serious, problems with the network.
Some users have sent trouble reports stating that their workstation was "slow". Others have issued trouble reports stating that their workstations kept printing one of the following error messages: le0: no carrier or NFS server not responding. Other trouble reports come in stating that inter- machine file transfers seem to take longer than they should.
All of these trouble reports are symptoms of network problems. In most cases, the network simply has "too many" machines connected to it. While the actual number of machines is not too large, the amount of information the machines want to send out on the network in a given period of time is larger than the hardware can handle during that period of time.
Through trial and error along with close monitoring, we have determined an acceptable number of workstations on each ethernet segment to keep the throughput demands within the limits of ethernet hardware. As we approach the acceptable number of workstations on a network, we "split" the network into two smaller segments. The ECN currently is comprised of roughly 35 ethernet segments which tie approximately 350 systems together.
Each segment is actually a small network. These small networks are tied together via "gateways". A gateway is one of the Sun servers, a Gould, or a Vax system. Most of the gateway systems are tied together by a network we call the ECN backbone.
Due to the number of trouble reports we received last Spring, the ECN backbone network was monitored to determine usage. We were watching to see the amount of data sent over the backbone network and whether it was approaching the point of being overloaded. Indeed, we determined that the backbone is quickly approaching the breaking point!
In a 24 hour period, we sent over 4 gigabytes of data through our monitoring equipment. Most of this data was sent between the hours of 2pm and midnight. In 48 hours, we caused the monitoring equipment to overflow its counters and display garbage data. The number of collision packets and other network errors were much larger than we ever expected. We decided it was time to do something to improve the backbone network.
Technology has finally come to the rescue of our aching backbone. The ECN has recently entered into an agreement with a company which produces boards for a fiber optic network system. This new network is called Fiber Distributed Data Interconnect (FDDI for short). FDDI is (minimally) 10 times faster than the ethernet we are running now. In the optimum mode, the FDDI network can run 20 times faster than ethernet. Besides being faster than ethernet, FDDI packet sizes can be 3 times larger than the ethernet allows. On top of this, a new network protocol (XTP) is being readied for the FDDI networks. The combination of faster media, larger packets and optimized protocols means that FDDI should be able to provide us with enough bandwidth for the next 5 to 7 years.
The ECN plans to set up an experimental FDDI network in the MSEE machine room during the Fall semester. This network will be used to develop software and help the manufacturer find and eliminate FDDI board hardware bugs. Once the network is running, we will put it into production for Fall semester load and get some live user test time on it.
After we have worked the bugs out of the new hardware and software, we plan to connect all of our "large" systems together via FDDI. This new FDDI backbone would eventually replace our ethernet backbone, allowing faster NFS file sharing, faster network dumps, and fewer network problems for many of our users.
Unfortunately, as with all good news, there is also some bad news. Currently, there is no way to run FDDI to every workstation that we maintain. The existing SUN 3/50, and 3/60 systems will stay on ethernet, as they do not allow the addition of FDDI hardware. Sun systems with a VME backplane (3/110, 3/140, 3/180, 3/160, 3/280, 4/280, 4/390, 4/110) will be capable of FDDI network links by adding the appropriate boards and installing optical fiber to the workstation locations. Other machines that will be FDDI capable are Gould NP1, Silicon Graphics workstations, MIPSco systems, Ardent systems, plus most any other type of systems that uses the VME bus for I/O controller communications. In addition, many manufacturers are working on FDDI hardware for their systems. We believe that SUN, and other vendors will soon be offering FDDI hardware for most of their product lines.
HTML