All pictures in the Cobalt Gallery appear here courtesy of Rochus Schmid. The rambling commentary is the fault of Serguei Patchkovskii.

You should realize that Cobalt consists of many, many pieces of hardware (end even more pieces of software - which, fortunately, take far less space than the creaking, groaning, glowing, and sizzling bits). It is therefore quite large. The room containing Cobalt is, on the other hand, quite small. Even the best wide-angle lens Rochus has simply can't take the whole thing in one shot. Until you can visit the machine room in person, you'll just have to trust me that the things are indeed like I say they are.

Clicking on each picture will get you a corresponding full-resolution image, at about 1800 by 1200 pixels. The full-resolution images are going to be fairly large - between 100 and 500 kilobytes each.

On the left picture is the close-up the the main console used to control all Cobalt nodes. It is a genuine Digital VT105 serial terminal - an almost antique piece of hardware of an immense rarity value. Guys from Digital were really impressed by it. The right panel shows me sitting at the console doing nothing. This is not the usual state of the affairs - for one thing, it would flare the tempers of the Cobalt Management Team, which can be seen in action on the picture below.

You'd already met me (on the left). The guy on the right hiding his seven-pronged whip is Tom Ziegler, who is the whole of the Cobalt Management Team - part time. He is also, or rather mostly a very successful theoretical chemist, which usually implies a deeply ingrained dislike of management problems of any kind.

Cobalt generates a not inconsiderable amount of heat. Each node requires between 100 and 150 watts of electric power, depending on the configuration and load. Most of this power gets converted to heat, so that the hardware closeted in the Cobalt machine room spews out about 12 kilojoules of heat each second. This is sufficient to boil a litre of tap water in under 30 seconds, and the things in the machine would have become really uncomfortable without the industrial grade cooling unit located at the back of the machine room. The nasty-looking tubes in the lower left corner of the picture connects the unit to the water mains. The water gushing out of the drains outlet (bottom of the picture) is usually quite hot - you do not want to accidentally put your foot under it!

Immediately to the left from the cooling unit is is the nerve centre of the Cobalt cluster, built of four 3COM SuperStack 3300 Fast Ethernet switches connected by a Matrix Module. The fifth unit at the bottom of the rack is part of the smaller brother of Cobalt - the Ozone. The pictures do not give the justice to the impressive array of the blinking lights on the front panel of the switch. In the real life they are much more appealing - although they are simply not in the same league as a mainframe console from the old times. A pity, really.

By examining the close-up of the right panel, you'll be able to see that all Cobalt nodes were online at the time the shot was taken - as indicated by the rows of cute green lights. Since we cannot see any yellow lights right above the the green status lights, all nodes were humming happily to themselves running jobs which had not required network communications at that particular time. The left picture (which unfortunately has the front panel slightly out of focus) was taken a few minutes later, and shows intense network exchange between the nodes 68, 77, 80, and 86. From the node numbers, I would guess it was a parallel ADF job which entered the communication phase just then.

These two pictures show a good fraction of Cobalt nodes. The right one was taken by Rochus standing right in front of the console (visible on the left picture), so that the nodes shown on it are in fact standing right across the aisle from the nodes shown on the left picture. As you can see, the boxes are not really sitting on benches, as you were promised on the Cobalt home page. Still, COBALT sounds a lot better than anything you can get from Computers On Shelves All Linked Together - and we can live with a little white lie as long as we don't have to find space for all these benches.

This picture is intended to introduce the very last member of the Cobalt cluster - cobalt94. If you examine the close-up of the sticker label you'll see that it proudly proclaims the node to be IT. Unfortunately, and as seems to be the general trend for the pictures of this kind, this photograph is faked. At the time it was taken, Cobalts with numbers above 40 had no stickers at all. Instead, Rochus took a snapshot of cobalt34 (as you can see from the IP address, and I can see from the serial number), and digitally edited the name. Still, it is a nice shot of the backside of an old-model Digital Personal Workstation 500au.

In case you start to wonder why it is the backside, and why all systems sit on the shelves with their faces turned to the wall - it is not because they've been misbehaving lately. The reason is simply that the serial port connectors (the two nine-pin male connectors in the top right corner) are located on the back of the case. Since I occasionally need to hook up a console directly to a node (for example if it had crashed, or for the initial configuration), these connectors have to be easily accessible. Unfortunately, the main power switch of each node is on the front side - so that switching the node's power on or off calls for a bit of simple acrobatics. It does not happen too often, however.

Unlike humans (and very much like goblins and their relatives kobalds, who gave their name to the chemical element Cobalt), Cobalt does not need any light, and is located in a window-less room. To get this picture, Rochus left some of the lights on in the back of the machine room, so it is not quite the real thing (which in fact looks quite dull and almost totally black). It is pretty close, however.

Remember that I promised you a guided tour of the Cobalt machine room should you ever come here? These three pictures were taken during the very first tour of the machine room for the members of our research group. That tour was conducted on the occasion of bringing the full complement of 94 nodes in the production use. If your browser supports client-side maps, this is also your chance to visit the personal home pages of some of our group members. Simply click on the outline of a person on one of these photographs, and you'll be taken directly to his or her home page. If your browser does not understand maps, or if the person you are interested in is not on any of the pictures, the Cobalt home page contains a complete list of all personal home pages on this server.

In the meantime, I hope that you enjoyed visiting the Cobalt Gallery. Bye, and see you back later.

Pictures shown on this page and linked from it were prepared for an output device with gamma=1.0. If colours appear too saturated on your computer, you may need to change the gamma setting in your browser or external viewer so that it matches the actual characteristics of your monitor. If your browser does not support gamma correction (Netscape does not), you may be able to get better results with an external viewing program, such as ImageMagic. You can learn more about gamma corrections at the Radiance home page.