STP path cost: I'd like a better answer

doctorlexus

I've been trying to find any mention of how IEEE came up with the revised STP path costs, but to no avail. For the original STP, path cost was just 1Gbit/s divided by link bandwidth. The calculation for RSTP is also fairly straightforward: 20Tbit/s divided by link bandwidth. But the revised STP path costs are given in a table with cost being 100 for 10Mbit, 62 for 16Mbit, 19 for 100Mbit, 4 for Gbit, 2 for 10Gbit. I can find no formula or algorithm for how IEEE arrived at these costs. Up to 16Mbit bandwidth, the path cost follows the old formula, but after that the values deviate from that simple formula.

They didn't pull those numbers out of a hat. They determined they were ideal by some means or method. Does anyone here know what it is?

Sy Kosys

https://supportforums.cisco.com/discussion/11784931/stp-bandwidth-cost-formula

The interwebs wrote:

The costs are recommended by the IEEE in section 8.10.2 the 802.1D standard published in 1998 (available for free at the IEEE website).... The 1998 standard allows for a 16-bit path cost value (held in software), 1 to 65535, and a 32-bit root path cost (the cost which is advertised in a BPDU field).
Its successor, 802.1D-2004, increases the path cost to a 32-bit value, providing far more granularity in assigning costs, and enabling the use of a static scale.
IEEE-SA -IEEE Get 802 Program - 802.2: Logical Link Control

I think this is what you seek. Here's also the link for the successor 802.1D-2004, relevant path cost info on pgs 152-154:

IEEE Standard Association - IEEE Get Program


Anyways, hope that helps!

doctorlexus

Thanks for the reply. I looked at pages 152-154, but most of the path cost information was for RSTP. If detailed information on revised STP path costs is in there, I wasn't able to find it with a cursory examination.

Techytach

[FONT=&quot]
[/FONT]http://packetlife.net/blog/2008/sep/5/spanning-tree-port-costs/

Most useful thing I could find. Short end is no one knows exactly.

Other than that I notice that it follows closely to the 10^8/bandwidth that OSPF uses but they add something else in.

Sy Kosys

Well crud....the best I've been able to find is that they kinda don't have an algorithm for the higher-bandwidths.... Found this:

CCNP R&S Switch 300-115 OCG, pg 158 wrote:

The original IEEE 802.1D standard defined path cost as 1000 Mbps divided by the link bandwidth in megabits per second. Modern networks commonly use Gigabit and 10-Gigabit ethernet, which are both either too close to, or greater than, the maximum scale of 1000Mbps. The IEEE now uses a nonlinear scale for path cost.

Yayyy nonlinear bandwidth guesswork!