Re: ISIS vs. OSPF, redux [7:4945]
>Hello all: > >Thank you for everybody who answered my previous IS-IS question. > >I have been compiling a preliminary list on the advantages and disadvantages >of IS-IS vs. OSPF, and where you might want to use one over the other. This >is what I have come up with Radia Perlman groupies like ISIS. John Moy groupies like OSPF. :-) > >IS-IS: >- has a foothold in ISP's for historical reasons, as it was developed >earlier than OSPF. Therefore, for backwards compatibility, ISP's continue >to demand routers that can do IS-IS It's not just historical; ISIS has evolved and been tuned to meet quite a number of ISP needs. Unfortunately, much of this is extremely poorly documented, either in RFCs or even in vendor manuals. ISIS routinely supports far more routers than OSPF in a single area, and many ISPs, at least in their core, are single areas. 1000+ ISIS speaking routers are routine. ISIS, especially with tuning, requires less bandwidth than OSPF, especially in the stable optical networks of carrier cores. >- also is used for out-of-band SDH management by telcos >- Converges slightly faster than OSPF I have seen no evidence for this. >- (Naturally) is multiprotocol, so can handle CLNS, and Decnet phase V >(which is CLNS) Remember at the time of its introduction, OSI vs TCP/IP was still a very open issue. In 1990-1991, when the first major ISPs were being implemented, CLNP support was an important business concern. Multiprotocol is far less of an issue these days. The world is moving to IP, with tunneling of legacy protocols. IPv4-IPv6 evolution is being planned quite carefully. >- Has some features that OSPF does not that can be useful in special >situations, like the OL bit, etc. Has a lead in the implementation of traffic engineering. I'm not sure that the OL bit or the equivalent OSPF database overflow feature have ever become that important, because iBGP wasn't well understood at the time of the design of both these IGPs. > >OSPF >- Is better known, and documentation for it is more readily available >- Has an overall richer set of features than IS-IS (at least, until the >latest IS-IS revisions) It's probably most precise to say that OSPF has the ability to have much greater control over what deliberately leaks between areas, and between the OSPF domain and other domains. This may be useful in ISP POPs, but is more important for enterprises that have distinct communities of interest (common applications). >- Is the standard link-state routing protocol for enterprises, and is also >popular in ISP's. Using the opaque LSA, is more flexible for uses that were not necessarily thought about at the time of development, which might very well become useful for things like sub-IP control. ISIS does allow the creation of new TLVs, but that has to be done at a code level. > >Does anybody have anything to add? > > >Also, I would like to know what people think the future of IS-IS is, >particularly after the latest revisions. Does anybody think that IS-IS will >be able to maintain and expand its foothold in ISPs, and even move to the >enterprise, or is it forever doomed to its niche (and why do you think so)? I'd hesitate to use the term "doomed." Large ISPs have different requirements than enterprises, so why should the same fundamental protocol be ideal for both of them? Things become much more complex when considering things like content providers. Both protocols have limitations and it isn't a given that significantly new versions, or even new IGPs, may come along. ISPs, in particular those that offer voice services, are very interested in subsecond convergence. That isn't going to happen without rethinking failure detection, and probably replacing the 40-year-old Dijkstra algorithm with a faster link state algorithm. Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=4972&t=4945 -- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
RE: ISIS vs. OSPF, redux [7:4945]
There are some design and operational advantages that IS-IS has over OSPF. As you mentioned, IS-IS is a ISO protocol which was originally designed to route CLNS networks. Because of this the CLNS topology is created using CLNS, not IP. Even when using IS-IS strictly for IP IS-IS still uses a CLNS based topology database with IP addresses listed as attributes to each CLNS entry. Due to this IS-IS does not require a complete SPF calculation when individual IP links flap. As long as the CLNS NET that the IP addresses are associated with do not change (flap or renumbered) then individual IP changes in the IS-IS database only require Partial Route Calculation and not full SPF calculation. A Partial Route Calculation consists of updating the appropriate IP information associated with a NET in the IS-IS database. This is less CPU intensive than full Dijkstra SPF calculations so single IS-IS areas have been known to scale much greater than OSPF areas. Depending on the network design, OSPF areas have been known to grow over 100 nodes in a single area while IS-IS has been tested at over 1000 nodes in a single area. This is a major reason why most service providers prefer IS-IS over OSPF. Also, IS-IS is not limited to a backbone area constraint as OSPF is. Therefore, there is no requirement to tie all areas to a single backbone area which allows more scalability in the network design as well as flexibility in controlling routes. There doesn't have to be a single backbone area with a route to all destinations. A drawback that relates to the absence of a backbone area within IS-IS is that all areas are treated as "stub" areas. This means that IS-IS routers within a area install default routes to all level 2 routers within their area with active level 2 adjacencies with other areas via the Attached bit. This can lead to sub-optimal routing due to the absence of specific routes to destinations however route leaking and other design methods can be used to overcome this... Cheers, -Michael Cohen -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of NRF Sent: Thursday, May 17, 2001 10:05 PM To: [EMAIL PROTECTED] Subject: ISIS vs. OSPF, redux [7:4945] Hello all: Thank you for everybody who answered my previous IS-IS question. I have been compiling a preliminary list on the advantages and disadvantages of IS-IS vs. OSPF, and where you might want to use one over the other. This is what I have come up with IS-IS: - has a foothold in ISP's for historical reasons, as it was developed earlier than OSPF. Therefore, for backwards compatibility, ISP's continue to demand routers that can do IS-IS - also is used for out-of-band SDH management by telcos - Converges slightly faster than OSPF - (Naturally) is multiprotocol, so can handle CLNS, and Decnet phase V (which is CLNS) - Has some features that OSPF does not that can be useful in special situations, like the OL bit, etc. OSPF - Is better known, and documentation for it is more readily available - Has an overall richer set of features than IS-IS (at least, until the latest IS-IS revisions) - Is the standard link-state routing protocol for enterprises, and is also popular in ISP's. Does anybody have anything to add? Also, I would like to know what people think the future of IS-IS is, particularly after the latest revisions. Does anybody think that IS-IS will be able to maintain and expand its foothold in ISPs, and even move to the enterprise, or is it forever doomed to its niche (and why do you think so)? I am especially interested in the opinions of people like Ms. Oppenheimer and Mr. Berkowitz. Thanx in advance FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED] Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=4976&t=4945 -- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
