z**r
发帖数: 17771 | 1 current mpls label is only 20bit long, so the maximum capacity is 1M, and
the Internet routing table increases so fast, it's approaching 300k now, and
it's very likely that in the near futrue the table could be larger than 1M,
what could be the solution to map the FIB to LIB/LFIB?
http://bgp.potaroo.net/ | s*****g
发帖数: 1055 | 2 MPLS label space has little to do with the size of internet, but more
related to how many BGP neighors a router can peer with (and VPNs it need to
provision), Vanilla IP over MPLS will need only one label per BGP speaker
for all BGP routes this speaker advertises. MPLS label is only locally
significant, it is highly unlikely a BGP route will have 2^20 BGP peers,
well, labels might need to be allocated for VPNs if BGP speaker is also MPLS
VPN PE, But still ... | z**r
发帖数: 17771 | 3 I think what you said is true for MPLS VPN label, but I am more talking
about the pure MPLS. My point is, currently most Internet routers especialy
the edge routers have the full Internet knowledge, and these routers are
pretty much all MPLS enabled. Tho the label is locally significant, the
router does need to assign a label to one FIB/RIB item and propagate it to
the upstream router.
I don't worry about the P/core routers coz they don't have to build up the
full Internet table.
to
MPLS
【在 s*****g 的大作中提到】
: MPLS label space has little to do with the size of internet, but more
: related to how many BGP neighors a router can peer with (and VPNs it need to
: provision), Vanilla IP over MPLS will need only one label per BGP speaker
: for all BGP routes this speaker advertises. MPLS label is only locally
: significant, it is highly unlikely a BGP route will have 2^20 BGP peers,
: well, labels might need to be allocated for VPNs if BGP speaker is also MPLS
: VPN PE, But still ...
| s*****g
发帖数: 1055 | 4 No, I was not talking about MPLS VPN label, I am talking about pure MPLS ...
The router only needs to assign labels for its IGP routes, it does not need separate labels for its BGP routes, it only needs to advertise ONE label for its BGP router ID.
Any router that has Internet routes installed in its IGP table, that network engineer should be fired.
especialy
【在 z**r 的大作中提到】
: I think what you said is true for MPLS VPN label, but I am more talking
: about the pure MPLS. My point is, currently most Internet routers especialy
: the edge routers have the full Internet knowledge, and these routers are
: pretty much all MPLS enabled. Tho the label is locally significant, the
: router does need to assign a label to one FIB/RIB item and propagate it to
: the upstream router.
: I don't worry about the P/core routers coz they don't have to build up the
: full Internet table.
:
: to
| z**r
发帖数: 17771 | 5 the BGP routes will be installed into IP table eventually right? anyway....
let me clarify the question from different point of view then, say the whole
network has IGP only, all 300k prefixes are IGP prefixes, and it's
approaching to 1M now
..
need separate labels for its BGP routes, it only needs to advertise ONE
label for its BGP router ID.
【在 s*****g 的大作中提到】
: No, I was not talking about MPLS VPN label, I am talking about pure MPLS ...
: The router only needs to assign labels for its IGP routes, it does not need separate labels for its BGP routes, it only needs to advertise ONE label for its BGP router ID.
: Any router that has Internet routes installed in its IGP table, that network engineer should be fired.
:
: especialy
| s*****g
发帖数: 1055 | 6 It is true BGP routes will be installed in FIB, but that has nothing to do
with LFIB space. Say one BGP router (say its BGP router id 1.1.1.1)
advertises 100K BGP route to a peer, for any one of those prefixes, the peer
will do IP lookup, and find out the next hop is 1.1.1.1, the peer will
further do recursive lookup for 1.1.1.1, and better you you have a label
switch path to 1.1.1.1 on this peer, so the look up will return something with a label. (RSVP or LDP), this peer just slap what ever lab
【在 z**r 的大作中提到】
: the BGP routes will be installed into IP table eventually right? anyway....
: let me clarify the question from different point of view then, say the whole
: network has IGP only, all 300k prefixes are IGP prefixes, and it's
: approaching to 1M now
:
: ..
: need separate labels for its BGP routes, it only needs to advertise ONE
: label for its BGP router ID.
| c*****i
发帖数: 631 | 7 if we have 300k igp routes say ospf routes, we need worry about ospf first...
I'm not sure about other edge devices, but for cisco, I don't think GSR/ASR
can support such large number of igp routes.
whole
【在 z**r 的大作中提到】
: the BGP routes will be installed into IP table eventually right? anyway....
: let me clarify the question from different point of view then, say the whole
: network has IGP only, all 300k prefixes are IGP prefixes, and it's
: approaching to 1M now
:
: ..
: need separate labels for its BGP routes, it only needs to advertise ONE
: label for its BGP router ID.
| z**r
发帖数: 17771 | 8 I know LDP doesn't allocate labels to BGP routes, but I thought it would do
after the best paths in the BGP routers are installed into the IP routing
table. good to know this.
Is there any further reading? Looks like a verdor proprietary implementation
?
I quickly checked RFC 5036, didn't find this tho
peer
with a label. (RSVP or LDP), this peer just slap what ever label it received
from its IGP neighbor for 1.1.1.1, this process is exactly the same
【在 s*****g 的大作中提到】
: It is true BGP routes will be installed in FIB, but that has nothing to do
: with LFIB space. Say one BGP router (say its BGP router id 1.1.1.1)
: advertises 100K BGP route to a peer, for any one of those prefixes, the peer
: will do IP lookup, and find out the next hop is 1.1.1.1, the peer will
: further do recursive lookup for 1.1.1.1, and better you you have a label
: switch path to 1.1.1.1 on this peer, so the look up will return something with a label. (RSVP or LDP), this peer just slap what ever lab
| z**r
发帖数: 17771 | 9 300k should be fine, I've seen 500k OSPF routes on Cisco routers
...
ASR
【在 c*****i 的大作中提到】
: if we have 300k igp routes say ospf routes, we need worry about ospf first...
: I'm not sure about other edge devices, but for cisco, I don't think GSR/ASR
: can support such large number of igp routes.
:
: whole
| s*****g
发帖数: 1055 | 10 LDP does not care where the prefix is coming from, all it does is FEC and
label mapping. Why LDP does not advertise labels for BGP routes? it is not a matter of proprietary implementation but rather a matter whether it makes sense or not. I am pretty sure you technically can make LDP to advertise labels for BGP routes.
There are different LDP implementations (ordered control vs independent control, liberal retention vs conservative retention), but that has nothing to do with the topic we are dis
【在 z**r 的大作中提到】
: I know LDP doesn't allocate labels to BGP routes, but I thought it would do
: after the best paths in the BGP routers are installed into the IP routing
: table. good to know this.
: Is there any further reading? Looks like a verdor proprietary implementation
: ?
: I quickly checked RFC 5036, didn't find this tho
:
: peer
: with a label. (RSVP or LDP), this peer just slap what ever label it received
: from its IGP neighbor for 1.1.1.1, this process is exactly the same
| | | w***s
发帖数: 321 | 11 500K or 50K OSPF Route? 没人这么疯狂吧，想搞死路由器没有比把BGP重分布到
IGP里更容易的事情了。
我倒是不担心MPLS Label空间不够，因为大家都在减少Label数量，就Cisco
而言，没有per vrf/router label前很痛苦。
【在 z**r 的大作中提到】
: 300k should be fine, I've seen 500k OSPF routes on Cisco routers
:
: ...
: ASR
| s**********9
发帖数: 1238 | 12
9494，应该是50K OSPF吧？
我记的OSPF一个AS内支持15000条路由就很勉强了，所以现在的ISP都用的ISIS做IGP，
也不过支持30K路由而已。
我记的04年的时候BGP路由大概200K左右，4年到了300K，离1M还远着呢，起码10年，到
时候肯定有新技术来解决！
【在 w***s 的大作中提到】
: 500K or 50K OSPF Route? 没人这么疯狂吧，想搞死路由器没有比把BGP重分布到
: IGP里更容易的事情了。
: 我倒是不担心MPLS Label空间不够，因为大家都在减少Label数量，就Cisco
: 而言，没有per vrf/router label前很痛苦。
| z**r
发帖数: 17771 | 13 I've personally done that many times, 500k is fine, cpu could have problems
depending on which type of routers, it may take minutes to converge
【在 w***s 的大作中提到】
: 500K or 50K OSPF Route? 没人这么疯狂吧，想搞死路由器没有比把BGP重分布到
: IGP里更容易的事情了。
: 我倒是不担心MPLS Label空间不够，因为大家都在减少Label数量，就Cisco
: 而言，没有per vrf/router label前很痛苦。
| z**r
发帖数: 17771 | 14 OSPF<->AS? what are you talking about?
yeah, ISIS is more scalable than OSPF tho
【在 s**********9 的大作中提到】
:
: 9494，应该是50K OSPF吧？
: 我记的OSPF一个AS内支持15000条路由就很勉强了，所以现在的ISP都用的ISIS做IGP，
: 也不过支持30K路由而已。
: 我记的04年的时候BGP路由大概200K左右，4年到了300K，离1M还远着呢，起码10年，到
: 时候肯定有新技术来解决！
| s**********9
发帖数: 1238 | 15
呵呵，我的意思是每个AS内的IGP路由目前ISP基本都是ISIS了，有3万条就足够了。
我想LZ可能看错了，500K条？那还要BGP来干吗？全用IGP就可以了嘛。目前你都查到
BGP路由有300K左右了，全换OSPF不好么？
就是因为IGP消耗大，支持不了那么多的路由，才有BGP的出现
链路状态协议的OSPF对CPU的开销比BGP要大多了。
实际上，50K的OSPF应该也没有的，我记的OSPF最多支持15000条路由
具体查查
【在 z**r 的大作中提到】
: OSPF<->AS? what are you talking about?
: yeah, ISIS is more scalable than OSPF tho
| s**********9
发帖数: 1238 | 16 OSPF最大支持路由条目，我一时间没查到。
但随便GOOGLE一下OSPF文档，都很容易查到，“合理部署的OSPF AREA 最多支持1024台
ROUTER”这句话
由此很容易判断OSPF最大的路由条目了 | z**r
发帖数: 17771 | 17 if you read the RFC, there is no limitation about how many routes can OSPF
support. 1k limitation is about the hops if I remember correctly
【在 s**********9 的大作中提到】
: OSPF最大支持路由条目，我一时间没查到。
: 但随便GOOGLE一下OSPF文档，都很容易查到，“合理部署的OSPF AREA 最多支持1024台
: ROUTER”这句话
: 由此很容易判断OSPF最大的路由条目了
| z**r
发帖数: 17771 | 18 BGP出现不应该是因为IGP消耗大吧，RIP消耗应该是很小了。OSPF消耗比较大，主要是
link state & flooding。
have you tried 500k? try it, I cannot say it's ok for any router, but should
be fine with 7600 and above.
【在 s**********9 的大作中提到】
: OSPF最大支持路由条目，我一时间没查到。
: 但随便GOOGLE一下OSPF文档，都很容易查到，“合理部署的OSPF AREA 最多支持1024台
: ROUTER”这句话
: 由此很容易判断OSPF最大的路由条目了
| w***s
发帖数: 321 | 19 那就只能停留在实验室阶段了。记得Light Reading测试过一次Edge router OSPF
performance, 当时的条件是50000左右的LSA，其中主要是Type 5, 其他的也就2K
以下。关键是没人拿自己的工作开玩笑。
MPLS Label的来源有几个：
1. LDP/IGP, 一般都设计成只分配PE RouterID label, 不该超过1K
2. TE/FRR, 大规模部署也就1k ~ 10k左右
3. VPN label，这个才是大头，不过如果支持per VRF label，也没多少。
所以安啦;-)
problems
【在 z**r 的大作中提到】
: I've personally done that many times, 500k is fine, cpu could have problems
: depending on which type of routers, it may take minutes to converge
| w***s
发帖数: 321 | 20 想当初有些人嫌EXP三位太少，没办法支持更多的QoS Class，还动脑筋到了Label上...
【在 w***s 的大作中提到】
: 那就只能停留在实验室阶段了。记得Light Reading测试过一次Edge router OSPF
: performance, 当时的条件是50000左右的LSA，其中主要是Type 5, 其他的也就2K
: 以下。关键是没人拿自己的工作开玩笑。
: MPLS Label的来源有几个：
: 1. LDP/IGP, 一般都设计成只分配PE RouterID label, 不该超过1K
: 2. TE/FRR, 大规模部署也就1k ~ 10k左右
: 3. VPN label，这个才是大头，不过如果支持per VRF label，也没多少。
: 所以安啦;-)
:
: problems
| z**r
发帖数: 17771 | 21 现实中目前是这样，不过这不是讨论的一个假设吗，呵呵
【在 w***s 的大作中提到】
: 那就只能停留在实验室阶段了。记得Light Reading测试过一次Edge router OSPF
: performance, 当时的条件是50000左右的LSA，其中主要是Type 5, 其他的也就2K
: 以下。关键是没人拿自己的工作开玩笑。
: MPLS Label的来源有几个：
: 1. LDP/IGP, 一般都设计成只分配PE RouterID label, 不该超过1K
: 2. TE/FRR, 大规模部署也就1k ~ 10k左右
: 3. VPN label，这个才是大头，不过如果支持per VRF label，也没多少。
: 所以安啦;-)
:
: problems
|
|
