Here’s a question: who uses Packet-Over-SONET interfaces and why the heck would you ever want to do that? At first glance, POS looks ridiculous because of the price. The following is a sample of the Cisco POS Shared Port Adapters available on the ASR1000 platform:
|SPA-2XOC3-POS||2-port OC3/STM1 POS SPA||$ 11.5k||
|SPA-1XOC12-POS||1-port OC12/STM4 POS SPA||$ 14.0k||601 Mbps||$ 23.29|
|SPA-1XOC48POS/RPR||1-port OC48/STM16 POS/RPR SPA||
|2405 Mbps||$ 15.80|
|SPA-OC192POS-XFP||1-port OC192/STM64 POS/RPR XFP Optics||
|9620 Mbps||$ 12.99|
By the time you get to the OC192 interface, the price-per-meg has finally dropped out of the stratosphere. If you are looking at POS you may be building an ISP backbone – so you’re probably deploying 10 Gbps links anyway. I work for an ISP, and we don’t build anything with less than 1 Gbps. Of course you’re dropping a cool $125k for a single interface.
In contrast, a 10-Gigabit Ethernet SPA (SPA-1X10GE-L-V2) is $2 per meg! Whoa. What gives? A 10GE interface runs at approximately the same speed as an OC192. Why would you pay 13X more for the same speed?
There are operational differences between Ethernet and POS.
Let’s say you have a /30 network provisioned across a PTP Ethernet link. You still need to know the MAC address of the device on the other side. That means you will have to perform an ARP. That might take a few hundred mili-seconds but you only have to do that once because the ARP cache should stay fresh as long as the link is seeing traffic.
POS uses PPP to encapsulate the IP traffic onto the SONET framing. As soon as the interface and the line protocol come up, it’s ready to go – no ARP needed. The advantage here is definitely with POS.
OSPF will handle an Ethernet link as a multi-access network. Each router on either end will flood a Type-1 LSA across the entire area to announce they know about the /30 transit network. The link-state ID will be the interface IP address of the designated router. There will have to be an election to determine who will be the designated router. Each router will also flood a Type-2 LSA that lists all routers on the segment (a grand total of two). So, that’s four flooded LSAs and one election that has to be dispensed with.
OSPF on POS acts as a true PTP link. Each router floods a Type-1 LSA advertising their knowledge of a transit network. No Type-2 is flooded and no election occurs. Your OSPF database will be smaller and your Dijkstra’s algorithm will run faster. Advantage: POS.
POS has real advantages over Ethernet, but does it justify the cost? I’m saying it doesn’t. A link may be revenue generating with lots of customers on it, but the benefits are too minor to add up.
Has anyone out there done any POS WAN links? What were your experiences? Did I miss something here? Feel free to chime in.