Spanning Tree Protocol

[bunker]

Fellas, I was wondering if someone can help me out here. I need to know if anyone out there in a metro ethernet or campus environment is running STP/MST/RSTP at a large scale. Large scale meaning about 200 switches. Can anyone point me to a link for a case study or a company doing this preferabley with Cisco gear but any gear would suffice.

Also, any links on STP scaling and free standards downloads?

I am searching myself but I recall this site has a tremendous amount of good folks.

Any help would be appreciated thanks..

[udunuwara]

I think it's not the no. of switches, but rather the no. of VLANs for instance. A ring running RSTP operating full 4k VLANs will probably take 600ms to converge if the root fails, according to some cisco guys. But this seems to be pretty high for me, but they say it's ok.But depending on your network, it could take seconds I guess.

standards available in the IEEE.org is for sale not for free.

[nrajanpanda]

STP as a standard recomends seven switches in a ring topology and RSTP may go up to 14 again recomended.There is no hard and fast rules as to how many switches to be accomodated in a ring.However my practical experience recomends the above. With RSTP typically the convergence is 600 msecs . This could be lowered by running multiple instances of STP in a ring through the use of VLANs (MSTP). In some cases it really can go up seconds unless the BPDY flooding accross rings or accross VLANs are not prevented through somemechanism.You can refer to cisco site for RSTP ,MSTP white papers. All metro providers (Reliance Infocomm,Tata telecomm.) are using MSTP in the rings with a maximum of 7 to 8 switches in a ring.

[pkokhan]

We run MST in a large DC. Previously we used STP but there were two problems. First is slow reconvergence. Second is that because we had too many VLANs and used trunks between distribution and access switches , the two distriburion switches had too many STP port instances (cisco recommends less than 4000) and CPU utilization reached 80-90 per cent.

I we moved to RSTP we would fix the first problem , not the second. So we migrated to MST. Mapped VLANs 1-2000 to instance 1 (root at sw1) and the rest - to instance 2 (root at sw2). It has been working perfectly for more than a year. CPU util is just 2-3 percent. I guess MST reconverges faster than RSTP if there are many VLANs, but not sure.

[j.samanta]

When we migrated from Rapid PVST+ to MSTP in a network of Cisco 3560G switches, we observed that all the link (FastEthernet) cost which by default is 19 have been increased to 200000. Any Idea on what basis this link cost are increased? Do MSTP use some different formula for link cost calculation.