re: Cisco's HFR Gets ModDistributed: "I can use a bunch of independent processors."
Modular: " I can uppdate BGP independently of OSPF."
SMP: I can place two or more processors on one board to handle the (non-distributed) CPU load on that board.
These are three totally separate concepts. Don't confuse them.
Additional separate concepts:
In-service upgrade: If a system is modular, it may or may not permit upgrade of a particular module without affecting other modules.
User modules: If a system is modular and/or distributed, it may or may not support user modules.
Control plane/data plane separation: High-speed routers must separate the control plane from the data plane. All of the above concepts relate to the control plane. if the architecture separates thje control plane fromt he data plane, it may or ny not be able to support each of the above concepts.
I have absolutely no experience with the HFR. However, it is clear that IOS does not implement any of the above concepts. SMP and separation are at least theoretically possible for IOS. The other concepts require a new software architecture.
re: Cisco's HFR Gets Mod<bold>Distributed: "I can use a bunch of independent processors." ... Additional separate concepts: ... I have absolutely no experience with the HFR. However, it is clear that IOS does not implement any of the above concepts. SMP and separation are at least theoretically possible for IOS. The other concepts require a new software architecture.</bold>
Incorrect. The Cisco GSR is a distributed system based on IOS. The PRP is a PowerPC running IOS. The linecards are MIPS based CPUs running IOS. They communicate at low speed over the backplane CANbus, or at high speed over the dataplane serial links.
re: Cisco's HFR Gets ModOf course, modular software has been around in class 5 switches for ages. Without being an expert in switched telephony, my impression was that it has proven not to be only efficent but outright necessary. So it is possible.
The HFR software modularity goes one step further than Juniper's, as some other people have noted. On paper, it seems like a great architecture. If it's gonna work or not, I dunno, we'll see. It's a bold step though, you gotta give it Cisco for that. It's by far the most interesting aspet of the HFR.
re: Cisco's HFR Gets ModI am afraid that I can not agree with you. To give a couple of examples that require control plane separation, IPv4/IPv6 dual-stack, unicast/multicast. Should an IPv6 problem affect your IPv4 traffic? Should a PIM crash bring down your IGP and BGP?
The argument will not work for v4/v6 dual stack except for ospf. Most of the other protocols are not designed to offer any seperation at all. It might have been good to do that, but IETF went in another direction. You could try things like seperating the update processing of BGP from the route processing of BGP, but I would not advise it.
PIM is a valid example. You could seperate PIM out from the rest, but the dependencies between PIM and other modules are still going to make independent upgrades difficult.
re: Cisco's HFR Gets ModIncorrect. The Cisco GSR is a distributed system based on IOS. The PRP is a PowerPC running IOS. The linecards are MIPS based CPUs running IOS. They communicate at low speed over the backplane CANbus, or at high speed over the dataplane serial links.
I dont agree with using distributed system in that way. What your describing is a hierarchical multi-processor IO system. That sort of thing has been around forever.
I would describe a distributed system as being something like the route processing itself being distributed among multiple processors with independent memory.
There is nothing particuarly difficult about doing a hierachical IO system in a router. Its well known how to do that. But a true distributed system along the lines of what caspian and pluris worked on is very complex and difficult.
re: Cisco's HFR Gets Mod- As far as memory protection, what good is BGP being up if OSPF is offline? One takes down the other usually. Anything that puts routes in the routing table can't really be well protected as a module.
depends on what you mean by offline. if the OSPF hits a bug and crashes then restarts, due to the graceful restart mechanism, the routing table may not see any glitch (assume that bug does not cause repeated crashes, such as slow memory leak, etc). Same for the bug fix for the OSPF without bring the router down and it can be non-stop forwording. BGP does not even need to know that.
re: Cisco's HFR Gets ModIt is 64 OC-192s in a 7 foot rack, so it is actually 1.28 Tb/s as per Cisco math.
As someone said, it is really for the captive Cisco customers like Sprint, SBC etc. who really will not buy any other router and do not want to pay for expensive router interconnects.
It is based on the same crossbar architecture so nothing has changed from a performance perspective.
Where does this leave the 128xx series :-) What with HFR, son of HFR and other thunder-rolling announcements, Cisco's customers must be really confused.
Modular: " I can uppdate BGP independently of OSPF."
SMP: I can place two or more processors on one board to handle the (non-distributed) CPU load on that board.
These are three totally separate concepts. Don't confuse them.
Additional separate concepts:
In-service upgrade: If a system is modular, it may or may not permit upgrade of a particular module without affecting other modules.
User modules: If a system is modular and/or distributed, it may or may not support user modules.
Control plane/data plane separation: High-speed routers must separate the control plane from the data plane. All of the above concepts relate to the control plane. if the architecture separates thje control plane fromt he data plane, it may or ny not be able to support each of the above concepts.
I have absolutely no experience with the HFR. However, it is clear that IOS does not implement any of the above concepts. SMP and separation are at least theoretically possible for IOS. The other concepts require a new software architecture.