[Beowulf] Jury rigged ethernet?
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
Robert G. Brown rgb at phy.duke.eduTue Jun 28 05:40:48 PDT 2005
- Previous message: [Beowulf] Jury rigged ethernet?
- Next message: [Beowulf] mixing MPI with other libraries...
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
One last remark. Continuing with google (google is your friend, after all) I encountered this: 10BaseT and [100BaseTX] has the capability to run at 10M bits/sec. [100M bits/sec.] for transmission over Shielded Twisted Pair (STP) cables or Un-shielded Twisted Pair (UTP) (100 ohm) cable which forms the Ethernet architecture. The devices belonging to this subsystem require interconnection via RJ45-type connectors and twisted pair cables (28 AWG minimum), and must conform to IEEE802 specification. This interconnect is point-to-point, commonly in a star topology. Normally this runs out to 100 meters. Normally configured in a star configuration. 10BaseT at a minimum uses "Category 3". 100BaseTX at a minimum uses "Category 5" minimum. The [nominally] 100 ohm cable impedance should be between 85 and 115 ohms and have at least 2 twists per foot. The total number of devices on a single chain is 2. 100Base-TX uses 2 pairs of Cat5 UTP or 2 pairs of 150 ohm shielded balanced cable as defined by ISO/IEC 11801. With 100Base-TX one wire pair is used for transmission, the other pair for collision detection. 100Base-T4 uses of 4 pairs of Cat. 3, 4, or 5 UTP in a star topology at a maximum distance of 100 meters per segment on RJ45 connectors 25Mbps per wire pair]. (Quote from: http://www.interfacebus.com/Design_Connector_Ethernet.html ) So apparently the ethernet specification allows for the use of STP explicity -- not really that surprising. The question then becomes what kind of STP is available in the space and whether or not it conforms with ISO/IEC 11801 (or near enough). And, perhaps, whether NIC and switch designers have been conforming with this part of the spec, given the paucity of STP-based cabling in anything but legacy situations. So 100BT on STP is almost "guaranteed" to work and would obviously be the cheapest alternative if it does. gigabit also has an STP extension -- I think its cabling specification is deliberately backwards compatible with 100BT. So this will almost certainly work as well if all four pairs are used, at least over a 10 meter run (which is short). To summarize, I think that the answer is: "Yes, you can run any of the 1 Gbps or slower ethernet segments over STP, depending on precisely what kind of STP you have available, as there are specifications on gauge, twists per meter, and impedance. Because of impedance matching problems and reflections, one should avoid mixing STP and UTP whereever possible, use care terminating cable runs (regardless of the wire type used). Short UTP patch cables are probably OK on the ends of an STP backbone run." rgb -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 189 bytes Desc: not available Url : http://www.scyld.com/pipermail/beowulf/attachments/20050628/b985a8d8/attachment.bin
- Previous message: [Beowulf] Jury rigged ethernet?
- Next message: [Beowulf] mixing MPI with other libraries...
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
