March 2001

 

Home
Why?
Log
Issues & Ideas
Products & Vendors
September 2000
October 2000
November 2000
December 2000
March 2001
July 2001
August 2001
September 2001
October 2001
November 2001

Revision 5: March 2001

  After a break to deal with other meaningless life stuff (like Christmas and work), the next rev of the rack came in March 2001.  All the major issues of the previous version were at least partially addressed.

The first major discovery (and subsequent addition to the network) was an effective backup solution, in the form of an Ecrix's VXA rakPak, a black 1U case that contains one or two SCSI VXA-1 tape drives. The Ultra SCSI drive has a native capacity of 33Gb, with hardware compression providing as much as 66Gb of storage on a single tape - and stinky fast! After some examination (and partial disassembly), I discovered that the 1U chassis is just a power supply and cable extender for the regular internal tape backup drive. But at this point I had to leave the backup in its own external chassis (I'm sure I'll find a use for the other unoccupied bay at some point, maybe a removable hard drive). The reason? 2U cases!

Antec makes a couple of styles of 2U case, the 2U20ATX250 was a perfect solution to the diminishing rack space problem - and only 20" deep! The challenge of a 2U case is that regular PCI cards won't fit anymore, the case is too shallow. A PCI riser card is provided with the case that allows two PCI cards to be installed horizontally.

With PCI only available, that meant all four servers would have to switch from AGP video cards to PCI video cards. In the case of Chef (the SQL server) and Stan (the web server), that was really the only change - the only other card in their chassis was the NIC. Both Cartman (the file server) and Kyle (the mail server) would be more challenging.

The challenge with Cartman was more cards - in addition to the video and NIC there was an Adaptec 29160 SCSI controller card, added to support the SCSI boot drives, DVD and now the tape backup. This was one card too many. The solution was the ASUS CUSI-FX motherboard, with video and LAN built in. Surrendering the ASUS CUBX-E motherboard meant losing the built in four channel (eight drive) IDE connectors, but that was easily remedied by adding a Promise Technology ATA/100 IDE controller.

Another issue with Cartman was the external drive array. Early incarnations of Cartman had extremely long (in excess of four feet) custom-made IDE cables that ran from the motherboard in the 4U case out the back and down to the 4U drive chassis. Even using the slow ATA/33 type cables, the external drive array was unreliable - the IDE cables were too long, and running ribbon cable outside the case just isn't smart.

With the new 2U case,  I tried to shorten up the run to the external chassis by removing the top of the 4U drive chassis and passing the cables through a hole in the bottom of the 2U case. For whatever reason, the Antec folks had decided to put a hole in the bottom of the case for a slide-mount drive frame. Completely disassembling the 2U case (metal filings are generally bad for electronics), a little work with a Dremel tool made the opening large enough for the IDE cables to pass through.

On top of that, I wanted to get better performance out of the drives, so I tried ATA/100 cables. ATA/100 is a touchy specification, highly dependent on the length and spacing on the cable. Officially the specification only supports cables 18 inches long, although 24 inch cables are readily available and apparently work fine. But 24 inch cables still weren't long enough. I managed to convince a local cabling manufacturer to make four 36 inch long ATA/100 cables, although the vendor provided no guarantee that they would work. The cables were just long enough, and the system booted fine, showing Mode 5 (ATA/100) access to the drives.

Unfortunately, within an hour or so, after steady use of the drive array (it is, after all, a massive MP3 store), Cartman actually hung and took a long time to recover with lots of drive errors. The ATA/100 cables are just too long to work properly. Had to find a better solution.

Meantime, Kyle (the mail server) was also a challenge to convert to a 2U case because it was a Slot 1 motherboard, rather than a Socket 370 motherboard - Slot 1 processors stick up higher, and won't fit properly into a 2U case. Replacing the Slot 1 motherboard with another ASUS CUSI-FX motherboard solved that, and provided the opportunity to rebuild the Exchange Server (which needed doing).

At this point I stopped - work and travel were absorbing too much time to finish rebuilding the servers, so they sat for quite awhile...

Problems with this revision:
2U conversion work not completed.
Needed a better solution to the drive array problem.