s*****g 发帖数: 1055 | 1 I am simplifying the dilemma I am facing:
Say I have a core switch, I need to establish 100+ OSPF adjacencies with
TORs, TOR will have point to point connection to core switch, how are you
going to address those point to point links?
1) option 1, put all links in a /24 network
2) individually have a /30 network
Pros of option 1), easy IP addressing, cons of option 1) each TOR will have
100+ 2WAY/OTHER adjacencies which will burden CPU
Pros of option 2) only see OSPF adjacency with core switch, cons: need to
create separate L3 interface for each link.
What do you think? think in big scale. When replying, don't argue why we are doing L3 to TOR. :-) | f*****m 发帖数: 416 | 2 option2.
如果你显每一个link单独分配一个L3 subnet麻烦的话, 一个丑陋但或许会省力一点点
的办法是 RFC5838
have
【在 s*****g 的大作中提到】 : I am simplifying the dilemma I am facing: : Say I have a core switch, I need to establish 100+ OSPF adjacencies with : TORs, TOR will have point to point connection to core switch, how are you : going to address those point to point links? : 1) option 1, put all links in a /24 network : 2) individually have a /30 network : Pros of option 1), easy IP addressing, cons of option 1) each TOR will have : 100+ 2WAY/OTHER adjacencies which will burden CPU : Pros of option 2) only see OSPF adjacency with core switch, cons: need to : create separate L3 interface for each link.
| s*****g 发帖数: 1055 | 3 I only looked RFC5838 headline, looks it is for OSPFv3, what does it do in
short?
【在 f*****m 的大作中提到】 : option2. : 如果你显每一个link单独分配一个L3 subnet麻烦的话, 一个丑陋但或许会省力一点点 : 的办法是 RFC5838 : : have
| f*****m 发帖数: 416 | 4 OSPFv3给IPv4. 一个好处是用IPv6 link-local address,这样你担心的L3 subnet分配
就不是问题了.
【在 s*****g 的大作中提到】 : I only looked RFC5838 headline, looks it is for OSPFv3, what does it do in : short?
| a**********k 发帖数: 1953 | 5 Agree, option 2 is good in terms of scalability.
【在 f*****m 的大作中提到】 : option2. : 如果你显每一个link单独分配一个L3 subnet麻烦的话, 一个丑陋但或许会省力一点点 : 的办法是 RFC5838 : : have
| s*****g 发帖数: 1055 | 6 Yeah, but how are you going to maintain those IP addresses? Any OSPF
developer can comment how many adjacencies (mostly 2way/DROTHER neighbors) a
modern high end L3 switch will safely handle? I will feel comfortable if it
can maintain 1000.
【在 a**********k 的大作中提到】 : Agree, option 2 is good in terms of scalability.
| R*****A 发帖数: 127 | 7 Always use point to point link for OSPF, never use multiaccess (ethernet) in
practice, for scalibitiy, stability, operation, and easy to trouble-
shooting. | z**r 发帖数: 17771 | 8 neithor is good, OSPF is not that scale, why not BGP?
have
【在 s*****g 的大作中提到】 : I am simplifying the dilemma I am facing: : Say I have a core switch, I need to establish 100+ OSPF adjacencies with : TORs, TOR will have point to point connection to core switch, how are you : going to address those point to point links? : 1) option 1, put all links in a /24 network : 2) individually have a /30 network : Pros of option 1), easy IP addressing, cons of option 1) each TOR will have : 100+ 2WAY/OTHER adjacencies which will burden CPU : Pros of option 2) only see OSPF adjacency with core switch, cons: need to : create separate L3 interface for each link.
| z**r 发帖数: 17771 | 9 in OSPF, it's not just how many neighbors the router/switch can handle, LSA
types, number of prefix, number of area etc. all play big role.
I had a situation that the PE router needed to build up approx. 500 sessions
with CE routers, each CE router fed in only about 20 prefixes or less, but
due to there were mixed LSA types, 3, 5, etc., the CPU on the PE routers
suffered a lot. So I had to provide 2 steps solutions, 1) make all these
areas totally stub or totally stub NSSA, this saved big immediately, but
could repeat the issue again if sites kept growing or a fat finger could
easily bring down the PE again. 2) migrate all the PE-CE OSPF to eBGP
So in your situation, 1000+ neighbors? hmm, what platform you are talking
about here? It's possible, but need to design and test carefully
a
it
【在 s*****g 的大作中提到】 : Yeah, but how are you going to maintain those IP addresses? Any OSPF : developer can comment how many adjacencies (mostly 2way/DROTHER neighbors) a : modern high end L3 switch will safely handle? I will feel comfortable if it : can maintain 1000.
| s*****g 发帖数: 1055 | 10 BGP is really not a good choice for high density connections here,
configuration overhead, poor ECMP support are showstoppers.
【在 z**r 的大作中提到】 : neithor is good, OSPF is not that scale, why not BGP? : : have
| | | z**r 发帖数: 17771 | 11 many ways to simplify the configuration, ECMP, hmm, does max path work? at
least bgp can scale to 2k+ neighbors easily on a mid end router
【在 s*****g 的大作中提到】 : BGP is really not a good choice for high density connections here, : configuration overhead, poor ECMP support are showstoppers.
| s*****g 发帖数: 1055 | 12 High performance cluster, only LSA1,2 and 3. out of 1K neighbors only 8-16
adjacencies(4-8 uplinks to different core routers) Stub/totally does not make much difference, each rack has one network, number of routes/LSAs shouldn't be a concern.
LSA
sessions
but
【在 z**r 的大作中提到】 : in OSPF, it's not just how many neighbors the router/switch can handle, LSA : types, number of prefix, number of area etc. all play big role. : I had a situation that the PE router needed to build up approx. 500 sessions : with CE routers, each CE router fed in only about 20 prefixes or less, but : due to there were mixed LSA types, 3, 5, etc., the CPU on the PE routers : suffered a lot. So I had to provide 2 steps solutions, 1) make all these : areas totally stub or totally stub NSSA, this saved big immediately, but : could repeat the issue again if sites kept growing or a fat finger could : easily bring down the PE again. 2) migrate all the PE-CE OSPF to eBGP : So in your situation, 1000+ neighbors? hmm, what platform you are talking
| w***s 发帖数: 321 | 13 how about PVLAN OSPF P2MP?
【在 s*****g 的大作中提到】 : I am simplifying the dilemma I am facing: : Say I have a core switch, I need to establish 100+ OSPF adjacencies with : TORs, TOR will have point to point connection to core switch, how are you : going to address those point to point links? : 1) option 1, put all links in a /24 network : 2) individually have a /30 network : Pros of option 1), easy IP addressing, cons of option 1) each TOR will have : 100+ 2WAY/OTHER adjacencies which will burden CPU : Pros of option 2) only see OSPF adjacency with core switch, cons: need to : create separate L3 interface for each link.
| s*****g 发帖数: 1055 | 14 Very good suggestion, unfortunately Arista EOS does not support P2MP OSPF.
【在 w***s 的大作中提到】 : how about PVLAN OSPF P2MP?
| w***s 发帖数: 321 | 15 说实话,P2MP也就是解决地址问题,真正能支持1000个邻居的除了BGP就是EIGRP |
|