s*****g
发帖数: 1055 | 1 For IP FRR, this is true because IGP will converge no matter what, but for
MPLS facility-backup FRR, since the path is pre-determined in the ingress,
ingress may not implement re-optimization of protected LSP, i.e, protected
LSP may or may not be adaptive, so traffic may stay in bypass path,
bandwidth requirement is not guaranteed especially N LSPs are riding on the
same bypass.
For detour FRR, bandwidth constrain is not a problem by nature. |
|
z**r
发帖数: 17771 | 2 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 | 3 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。 |
|
s**********9
发帖数: 1238 | 4
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！ |
|
z**r
发帖数: 17771 | 5 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. |
|
w***s
发帖数: 321 | 6 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 |
|
w***s
发帖数: 321 | 7 嘿嘿，C家的产品我不知道的还不算太多，无论是J的RE或者是C的RP，只要是同
LC Chassis的往一块连的，跑个IGP/BGP的，同传统的产品就没什么不一样，连
程序都不需要大改，对于这台路由器而言都是Internal的，不论是放在何种Chassis
里面，隔着有多远。 |
|
w***s
发帖数: 321 | 8 来自主题: EmergingNetworking版 - 问个俗问题 我知道还问么。
说的就是这个老掉牙的俗问题，只是O派要举例的时候就是只有ATT了。
转flat IGP结构的时候，好多都选择了O->I迁移。 |
|
z**r
发帖数: 17771 | 9 why? the igp label distribution should be the same bah? I made the send-
label method work for inter-as l2vpn, and at the same time, the inter-as
l3vpn works fine too. the ebgp between ASBR's should do the same job
for |
|
z**r
发帖数: 17771 | 10 sometimes the IGP isn't consistent throughout the network, say 2 companies
just merged together so there might be some cases that you may use bgp send-
label to distribute labels
have
be |
|
c*****i
发帖数: 631 | 11 yes you can. option c is more about using RR to send vpnv4 label. some isp
even use igp with option c.
C) |
|
z**r
发帖数: 17771 | 12 pe3到ce3是ethernet access，pe3到upe是mpls access的h-vpls，在pe3上，需要把对
upe的split-horizon关掉，在upe上就是一个对pe3的software mode eompls。
pe3和upe需要有end to end lsp，所以平时如何setup这个eompls，这里也一样，比如
首先igp converged，然后turn on ldp in the cloud between pe3 and upe，最后就
是一个targed ldp between pe3 and upe for eompls |
|
z**r
发帖数: 17771 | 13 现在LDP只能work with IPv4，也就是说，即使跑6VPE了，那么P routers仍然跑的是
ipv4 IGP，否则LSP不能建立。所以比较好奇那些pure IPv6 network是怎么处理这件事
情的。貌似spec for LDP over IPv6已经有了 |
|
t*******r
发帖数: 3271 | 14 没听说过BFD和GR/NSR一起用出过什么问题, 唯一需要注意的是BFD如果是下到PFE层面(
delegate-processing, post version 8.5), 你要注意一般设为timer>100ms.
BFD:切换检测工具, 毫秒级
GR:只要helper可以, 你的流量不必中断
NSR:自己玩自己的,对方咋样都无所谓
BFD打开无非是检测你的IGP/BGP nbr down的快了一些而已, 除此之外还有啥?
只要查到NBR DOWN,就启用GR/NSR呗.
注意对JUNOS来说比较怪异的协议(如MVPN等)还是不要打开NSR, 个人感觉做的还是不太
完善. |
|
z**r
发帖数: 17771 | 15 mpls不在乎是什么routing protocol啊，只是如果你需要做TE，那么得是link state
IGP。
TE的确可以不要MPLS在core里跑，只要做full meshed PE-PE TE tunnels，就可以，当
然PE上还是要enable mpls的 |
|
s*****g
发帖数: 1055 | 16 So you are saying that if all core routers are running iBGP full mesh, then
you don't need IGP? Then we are talking about the same thing, and I am
telling you that you are wrong. ;-)
Of course you need to configure MPLS in the core (define configure MPLS?), TE can establish the LSP you want, but you still need MPLS functionality to switch the labeled packets.
Guys, guys, networking 102
mesh。 |
|
c*****i
发帖数: 631 | 17 full mesh bgp，bgp neigbour之间都是直连的，为啥要igp，你自己去试一下就知道了。
另外te是靠rsvp setup的和mpls ldp没有关系。我可以肯定的告诉你，而且有isp这样
用。
then
TE can establish the LSP you want, but you still need MPLS functionality to
switch the labeled packets. |
|
s*******8
发帖数: 12734 | 18 是不是ibgp必须要有一个IGP在下面，才能运行啊？
或者，直接directly connect才行？ |
|
|
s*******8
发帖数: 12734 | 20 嗯。做完了。
关键我IGP没学扎实，所以问问题抓不住关键。
很多东西就是一着急发现很烦，事后看过去，思路很清楚啊。豁然开朗啊。哈哈。 |
|
s*****g
发帖数: 1055 | 21 I don't know what is the best OSPF book, but
"The Complete ISIS Protocol" by Hannes Gredler & Walter Goralski
is a must read.
RIP is barely qualified as an IGP, EIGRP is widely used by some loyal Cisco enterprise customers only so you can probably ignore it or spend time on it in your leisure time. |
|
z**r
发帖数: 17771 | 22 肯定低，mpls core虽然可以代替atm了，但是还是需要sonet设备啊，再加上IP设备。
mpls-tp则直接把这两种合在一起，而且应该不需要维护IGP，无论设备成本还是运营成
本都低 |
|
d****i
发帖数: 1038 | 23 对， 要求就是没有IGP的情况下mpls还能跑起来，但是哪个哥们在配置的时候搞错一个
小地方就得查半边才能搞明白。
我觉得有点象钓鱼工程啊， 现在说得这也好那也好，真用起来了需要更新这个更新哪
个， 最后成本不一定低。哈哈
对了，7600马上要支持TP了。 7600， 6500这些老同志还真的不能退下来啊。 |
|
b******s
发帖数: 5329 | 24 语音最近几年没搞了。sp环境单一，每天就是igp, bgp, mpls，vpls, label,tunnel,
再不就是76, asr, crs,linecard, RSP, MX，T。。玩段时间就累了。 |
|
z**r
发帖数: 17771 | 25 看到一个文档，说在做BGP RR的时候，如果IGP是ISIS，并且interface MTU不是使用
default MTU，那么推荐使用clns mtu 1497.
有人明白为啥是1497而不是更大或者更小？ |
|
z**r
发帖数: 17771 | 26 neighbor reachable time设成多大，就在那个interval 有packet drop。这个怪俺第
一个帖子没说清楚，只笼统的说了一句nd timer，应该是nd ReachableTime。
俺是通过两个办法fix这个的，第一种去掉static，跑OSPFv3 or ISIS，第二种是把
static routes的next hop改成对方router的link local address。不过说到底，这两
种其实还是一个fix，dynamic routing protocol(IGP)也是用link local作为next hop
的。
那问题是为什么用link local就可以fix呢？ |
|
w***s
发帖数: 321 | 27 这里涉及到VM的定义，广义一点包括System VM和Process VM，后者有代表性的例子就
是Java VM。但是平常大家谈论得最多的还是狭义的以VM Ware为代表的System VM.
ASR1K刚出来的时候的VM HA让很多人困惑，最后PM承认XE中，IOS是作为应用程序跑在
Linux Kernel之上的，新版的图片也不提VM一词，代之以process。考虑到IOS本身的单
进程特性，有上下文说明的情况叫个Process VM也无不可，但是如果没有说明，还是不
用VM为好。
XE中Linux主要取代了原IOS的Host功能，负责加载管理ESP image/SIP image和IOS系相
关进程，同时提供IPC功能。
IOS系包括了从原IOS剥离出来的chassis manager，fwd manager，SPA driver，以及一
个完整Control Plane: IOSD。
在RP/FP/SIP上主要XE进程:
22578 ? S 255:20 | \_ /tmp/sw/mount/asr1000rp1-rpcontrol.03.01.
0... 阅读全帖 |
|
a***n
发帖数: 262 | 28 Teach you one trick -)
print configuration of your devices, one by one
read all of them and figure all them out.
You can focus on one thing at a time
such as IGP, BGP, Security, .....
Now you should be very familiar with your current
setup and configuration.
Print all CiscoLive pdf related to your area,
read all of them, several iterations, then you
should have a better understanding.
If not cisco, fine, read configuration guide |
|
t*******r
发帖数: 3271 | 29 从IGP分label折腾到3107的目的是啥?
不就是让路由条目并不随着VPN路由条目增长嘛
路由条目倒是可以随着PE的数量的增长而增长. |
|
|
|
s******v
发帖数: 4495 | 32 这个正常，bgp一半用在，between AS，不像IGP, within AS，所以应该的 |
|
d******t
发帖数: 834 | 33 nope.
如果没有冗余链路，建ebgp邻居最好用直连接口。
真要用loopback建，要互相添加静态路由的，ebgp peers之间没igp，network有啥用。
。。 |
|
x*********n
发帖数: 28013 | 34 那loopback这个subnet不在RIB上，怎么让router知道呢？
要redistribution到IGP上么？ |
|
s******v
发帖数: 4495 | 35 ebgp应该是between as，之间应该是没有igp的。在edge上，如果是peering with
loopback,就是static。当然在lab里面就无所谓了。 |
|
t*******r
发帖数: 3271 | 36 LZ应该把BGP table全导到IGP里看看结果.... |
|
s******k
发帖数: 234 | 37 民工确实比较闲。。嘿~~~现在做实施。。没项目的时候自然就闲啦。。。
我2010年左右在国内最大宽带运营商（不说名字是怕搜到麻烦）的集团IPv6项目组，这
个项目是涛哥过问，中央拨一部分款，是一个亿级的战略研究项目。项目每周直接
Report给董事长。当然项目组的研究内容不会仅仅是一个小小的IPv6。当时的规划是4
个省（上海，广州，江苏，四川）的部分区域上IPv6，现在上海，深圳，无锡，成都应
该已经有了大量的IPv6客户了。我10年脱离项目组最后一个会就是某个城市的具体V6设
备采购问题了。这些城市然后加上世博和深大运会做公开试点。目前国内的IPv6用户我
估计吧，可能超过100万了(这个纯粹拍脑门想出来的)，不过这100万人大部分都是在不
知情的情况下。目前的IPv6测试模式都是用户一次拨号同时获得IPv6/IPv4地址。通过
这种双IP的模式逐渐过渡。确确实实没有什么主动性推动。后面的消息我就不太清楚了。
我觉得ATT没有可能是因为他们根本就不急迫，因为他们有大量的存货。但是我们没有。
现在IPv6最大的问题就是。。。以往技术都是由应用推动技术，技术再催生应用这样一
种模式存在... 阅读全帖 |
|
s*****g
发帖数: 1055 | 38 PFPF, educate us, more detail please.
If I had the opportunity to re-achitect my company's network infrastructure,
I would never run IGP over WAN, I would use BGP if all possible, BGP is so much more flexible, scalable, less overhead and converges much faster. |
|
f*****m
发帖数: 416 | 39
finer region designation, such that people, say from Santa Clara county,
will go to a DC in Santa Clara, not to the DC is LA which servers west coast.
如果别的条件都一样的话, IGP cost应该会作为条件,这样Santa Clara的DC还是优先.
suck, more importantly, it is hard to design fall back, because routing is
mostly out of our control. We have to do geo-aware DNS at the same time with
BGP anycast. Each region has its own anycast range, we now have to have
finer regions, hence more anycast ranges. |
|
s*****g
发帖数: 1055 | 40 We send communities for ISPs to set preference, IGP cost is way down in BGP's decision chain, besides, we are peering with different AS in different locations.
have
coast.
with |
|
f*****m
发帖数: 416 | 41 如果是cost-community的话,还是在IGP的后面. 你们用自己的ASN 还是SP的? 如
果用SP的,自然是本地的优先啊.
BGP's decision chain, besides, we are peering with different AS in different
locations. |
|
s*****g
发帖数: 1055 | 42 Not necessarily, according to the article, Google is deploying openflow on
its WAN infrastructure in stead of data centers. One big limitation with
routing protocols (IGP or BGP) is that only best route ("best" in general
term) is selected, i.e. it is hard to implement traffic engineering, let alone to implement application aware traffic engineering, PBR, BGP and MPLS TE implement traffic engineering in a far from elegant/scalable way, I
assume OpenFlow does not have this limitation.
， |
|
l***y
发帖数: 791 | 43 In the rfc it says mpls forwarding state means either in-lable/outlabel-
nexthop for non-ingress or fec/out-label-nexthop for ingress, but what is '
next hop' in this context? Interface or next hop neighbor ip?
So if IGP updates but for some reason nothing updates ldp, i.e. if ospf
routes around a router with dead RE because ospf timed out but nothing else
has gone down (ldp has much larger hold timer), would mpls forwarding still
happen towards this router with dead RE?
Also, i'm reading about ... 阅读全帖 |
|
s*****g
发帖数: 1055 | 44 The same problem exists between IGP and BGP, what was the solution for that?
It makes a difference but the most vendors implement ordered control with
downstream unsolicited. i.e, a LSR only advertises a label for a FEC (
usually /32 loopbacks by default but other FECs can be injected by
configuration) upstream when a FEC is egress or when it has a mapping for
that FEC. |
|
l***y
发帖数: 791 | 45 i would think igp/ldp sync has somthing to do with size of LIB. LIB != LFIB.
/ |
|
z**r
发帖数: 17771 | 46 not neccessary, igp won't come up if the ldp adjacency is not up
LFIB. |
|
s******v
发帖数: 4495 | 47 如果是NH的话，在ipv6 pfx, NH = ::FFFF:x.x.x.x，这个解析应该是从IGP(ospf in
my case)得到啊。
可是为啥要在BGP里面的IPV4 AF里面，“nei x.x.x.x act" |
|
z**r
发帖数: 17771 | 48 NH应该是far end PE loopback，这个肯定是从IGP得到，同时也应该在LFIB里，你不激
活ipv4 neighbor，同时看一下show mpls for？
刚才俺也做了个实验，果然不需要在add ipv4下激活ipv4 neighbor，只需要在add
ipv6下激活ipv4 neighbor就可以送ipv6扩展了。
R1----R2----R3----R4
R1 and R4 are dual stack, R2 and R3 are IPv4 only, LDP enabled on all
interface
相关配置如下
!
!
!
ip cef
no ip domain lookup
ipv6 unicast-routing
ipv6 cef
!
!
interface Loopback0
ip address 10.1.1.1 255.255.255.255
!
interface Loopback1
no ip address
ipv6 address 2001::1/128
!
interface Ethernet0/0
no ip address
... 阅读全帖 |
|
x*********n
发帖数: 28013 | 49 大侠，这个signaling指的是啥啊？能在指点一下么？
你是说IGP protocol一样，需要Hello传递一下信息的package么？ |
|
l***y
发帖数: 791 | 50 Got answer:
Limitation of VRF-Lite:
PE <-> CE running VRF-lite
|_____ if this link is traditional WAN then it requires multiple WAN
circuits or Frame-relay encap.
Alt. solutions of Spe/Upe:
multiple ospf area in IGP?
split mpls domains and have option A at split point?
UPE/SPE is kinda cumbersome to implement/roll-out/troubleshoot. Also bonds
us to Junos more.
I am no expert so just taking notes. Zher has comment? All comments welcome!
RR
draft |
|
