CERN Accelerating science

This website is no longer maintained. Its content may be obsolete. Please visit http://home.cern/ for current CERN information.

next up previous
Next: DECnet Terminates to INFN and UK Up: cnl230.html Previous: Software Development Tools Service

Computer Networking at CERN

  Mike Gerard IT/CS


Will you please help us to improve network performance?

Why should we need to ask you?

Over the last few years there has been an explosive growth in the use of the CERN General Purpose computer network and the number of systems which are connected to it. The inevitable result is that it is no longer possible to have a central support group which does everything for everyone: manpower requirements for such a service are simply not available. Thus, as with creating documents, end users have to become directly involved to some extent.

Another reason is that modern PCs and workstations have increased in power inexorably fast. In order to be able to handle this increase the CERN physical network has been modernised with the introduction of structured cabling. With this introduction every office has several network outlets, each capable of handling a connection at high speeds (100 megabits/second or above).

This said, the actual performance as seen by the office system depends not only on the user's office equipment but also upon the network equipment to which all of the network outlets are connected. This network equipment is normally a 10 megabit/second hub, on which the network access is shared between all of the connected users. Thus, most office users are being provided with a connection capacity of 10 megabits/second and shared with 20-40 other users. This is already a great improvement on the previous situation, and in most cases is sufficient for the average traffic quantity generated. There is, of course, much talk about faster networks - 100 megabits/second, even a gigabit/second - and IT/CS Group is very active in obtaining and testing this new technology on the installed structured cabling.

Therefore, the situation today is that the structured cabling has given us the possibility to offer real high-speed networking to most of CERN. However, most network users are still sharing 10 megabit/second network segments with many other people and yet have the computing performance to use all of the available capacity of their shared segment. The conclusion from this is that computer users in CERN, and even outside, should help ensure a fair usage of available resources.

The rest of this article details ways in which you can provide such help.

A brief note on the new structured cabling

A few years ago all of the general networking at CERN was done using coaxial ethernet cables linking a chain of computer systems together. Because any errors in, or breakage of, the chain could have disastrous effects it was decided to re-wire most of CERN with so-called structured cabling. In this scheme every office has a number of network outlets which are connected to a single star point in the same building or in a nearby one. Inside the star point there is then a patch cable which can connect outlets that are required to be operational to a particular network hub. Each hub (or pair of hubs) may offer a particular service type.

There are different types of service offered, according to the needs of the computer system concerned, and this is done by connecting the computers to multi-user hubs which are in turn connected to routers. The routers ensure that on any given type of service only the network protocols required by that service are permitted, thus eliminating unnecessary traffic.

The introduction of routers normally follows some weeks after the installation of the structured cabling. Many users will have noticed this introduction because of having been required to change the configuration of their system(s). The easy way to know if routers have been already introduced is to check if the Internet Protocol (IP) Address of the computer system is of the form 137.138.x.y, rather than the previous 128.141.v.w.

We need to know what equipment is connected to the network

Yes, we really do need to know the details of what equipment is connected onto the network, where it is, what it is, whom it belongs to and so on. This is not just silly bureaucracy, but a real need allowing us to identify any network-connected equipment which is causing problems, or even suffering because of others, and explain the problems to the named Contact Person and Main User. Without such information all that we can do is to block network access to the problem system, when it can be found, which pleases neither us nor the real owner.

You should note that ALL equipment which may be connected to the CERN network should be registered, including portable equipment and equipment not needing any IP number.

You may find it very revealing to use the Web to check on equipment in your own area. We recommend that you open the Web page http://network.cern.ch/register and then select the option/operation:

In this page you can choose to search using various criteria. As well as looking at specific equipment by name we would strongly recommend that you look for all equipment registered for your name and for all equipment in your office. We guarantee that many people will be surprised by some of the results!

Why not do it now?

When you need to move equipment around!

Those of you that follow the above suggestions for checking on your network equipment may quickly realise that we have much wrong information in our equipment and cabling data bases. In part this is because it has been customary to move equipment around inside CERN without informing anyone. Unfortunately, this is no longer possible! Even in a single office, where all the network outlets may look the same, it is quite likely that moving an equipment from one outlet to another will mean that it no longer works properly, perhaps not even at all.

To permit the updating operations needed when moving equipment around we have included in the Web form (http://network.cern.ch/register) specific tools entitled:

To avoid problems it is essential that these forms be used : the subsequent action, which may involve re-patching connections in a star point, will normally be done inside a few working hours.

Computer networking internal to CERN

As already stated, even with the structured cabling most users are in a situation where they are sharing an ethernet with around 20-40 other users. This is already a great improvement on the previous situation of users sharing with up to several hundred other users with very little protection against faulty hardware or software elsewhere.However, the ethernet segment is still limited to 10 megabits/second total throughput even if the ethernet interface in the user's equipment is capable of supporting the 100 megabits/second Fast Ethernet.

If this sharing and/or speed limitation becomes impossible to tolerate then IT/CS Group can advise upon ways to improve the situation, subject normally to the users requesting the improvements having a budget available to pay for such improvements.

Where no improvement is immediately possible there are a number of simple rules which, if followed, can help considerably:

Access to the external world

Although network communications to the rest of the world are improving they remain both costly and offer much less capacity than the network inside CERN. This is particularly true of the communications to the USA, to where we have just a 2 megabits/second line. However, the explosive growth in Web traffic, plus the need to transfer large files of raw physics data to various USA Laboratories, means that this line can easily be saturated. In fact, the statistics, which can be seen on-line at URL:

http://sunstats.cern.ch/mrtg/cernh8.html

confirm that the line often is saturated.

Again, there are a number of simple rules to help keep this USA link situation reasonable:

Our actions in case of observed network problems or excessive usage

Despite our best efforts, and those of the great majority of users, there are times when our overall monitoring indicates that there are either problems or some excessive use of network facilities. In such cases we will always try to contact the person(s) responsible for, or affected by, these problems or excessive usage. Our experience is that those responsible for such systems are always willing to attempt to track down the causes of the problems and eradicate them.

If you think that there are network-related problems

As an end user you may sometimes have felt that some task that you are trying to perform, and which requires network traffic, is operating much more slowly than usual. You may often feel that it is due to a network problem - some PCs with Microsoft Windows have been known even to say this (often wrongly!). We, as network operators, know that in many cases there is absolutely no network problem, but we also know that such a reply would not solve your problem. Therefore we are always willing to try to help identify the real problem.

You, in turn, can help us in a number of ways. Try to be quite clear on which is your own system, to which other system is it trying to communicate and for what purpose. You can also help by checking if any other nearby people have similar problems.

The information that you gather can be collected and sent to us by electronic mail to netops@cern.ch : this is what we would prefer if the problems are not critical. If they are critical then you can call us at 74927. In the latter case remember that if lots of other people are seeing the same problems then we may be forced to leave the phone to accept recorded messages only while we deal with the problem.

If you want to know more

There is a large amount of information available via the Web pages of CS Group, specifically including more detailed information on structured cabling and the Routing and backbone project

http://network.cern.ch

so feel free to browse as much as you like (or as much as the network will let you!). In cases where this information is insufficient, or if you wish to know more about the possibilities for improvement of your own part of the network infrastructure, please feel free to contact either Network Operations (netops@cern.ch) or myself (Mike.Gerard@cern.ch),


next up previous
Next: DECnet Terminates to INFN and UK Up: cnl230.html Previous: Software Development Tools Service

Cnl.Editor@cern.ch