n**********l 发帖数: 271 | 1 Is it possible for discontiguous networks (for example, multiple datacenters
across the world) to use the same ASN? A private backbone connecting them
is a must?
Route reflector does not solve the discontiguous problem, does it?
I was asked a bunch of questions for a CDN network and last night I realized
everything I said would make no sense if it has a noncontiguous backbone.
AS65000 East coast datacenter -- AS 100 (Upstream1) -- AS65000 West coast
datacenter
AS65000 East coast datacenter -- AS 200 (Upstream2) -- AS65000 West coast
datacenter
Without a private backbone, blackhole is very likely to happen. And even
with a private backbone, hauling traffic around is expensive and it is
essentially someone else's (transit providers etc.) work.
How do CDNs(Akamai, Netflix etc.) use BGP? Route reflector/modifying next-
hop?
I am quite confused. | s*****g 发帖数: 1055 | 2 A basic BGP feature "bgp allow-as in" will allow prefixes with its own AS in
AS-PATH into local BGP/routing table, by-pass BGP's loop-avoidance
algorithm, so to speak. Is this what you are asking? or if you are not multi
-homing, you just import default route from your local ISP.
Unless you are tier-1 ISP, you most probably don't need your own transport.
For Netflix's CDN, read their "open-connect" white paper, nothing fancy at
all.
datacenters
realized
【在 n**********l 的大作中提到】 : Is it possible for discontiguous networks (for example, multiple datacenters : across the world) to use the same ASN? A private backbone connecting them : is a must? : Route reflector does not solve the discontiguous problem, does it? : I was asked a bunch of questions for a CDN network and last night I realized : everything I said would make no sense if it has a noncontiguous backbone. : AS65000 East coast datacenter -- AS 100 (Upstream1) -- AS65000 West coast : datacenter : AS65000 East coast datacenter -- AS 200 (Upstream2) -- AS65000 West coast : datacenter
| x*********n 发帖数: 28013 | 3 你要从头理解它。
AS本来就是一个virtual的concept,和area一样。
现在你几个location是一个company,你用什么iBGP还是eBGP?
iBGP require direct connection吗?为啥你config iBGP的时候不用改TTL?
后面的问题你应该都知道的,那就是自问自答了,呵呵。
再说RR,为什么要RR,是因为full mesh database大,那跟next-hop有什么关系? | n**********l 发帖数: 271 | 4 一个interview BGP diagram里面sites之间没有link, 直接连upstream
我就看得莫名其妙的了....估计答的大都错了....
【在 x*********n 的大作中提到】 : 你要从头理解它。 : AS本来就是一个virtual的concept,和area一样。 : 现在你几个location是一个company,你用什么iBGP还是eBGP? : iBGP require direct connection吗?为啥你config iBGP的时候不用改TTL? : 后面的问题你应该都知道的,那就是自问自答了,呵呵。 : 再说RR,为什么要RR,是因为full mesh database大,那跟next-hop有什么关系?
| x*********n 发帖数: 28013 | 5 eBGP needs to directly connected, iBGP doesn't
eBGP 不directly connected也能peer,3种方式,一种是常见的emulti-hop,一种是
disconnected,一种是TTL-max啥的,总之都是改TTL。
BGP是over TCP的,你想啊,这都是在layer 4了,底层是要有IGP running的。
synchronation都是要和IGP match才能做BGP的。
你可以把interview题目发上来,大家帮你看看,弄清楚了,下次就不会错了。
【在 n**********l 的大作中提到】 : 一个interview BGP diagram里面sites之间没有link, 直接连upstream : 我就看得莫名其妙的了....估计答的大都错了....
|
|