n**********l
发帖数: 271 | 1 How can it be done on a Cisco/Juniper? Why would people do that except for
testing/troubleshooting?
1. use weight(Cisco)/preference(Juniper), does it cause problems with iBGP
since other BGP routes don't know about this (multiple peering locations)?
2. change local-preference with route-map? Does it scale? Do it on all edges
for all prefixes?
And why in the world would you do that? Maybe one link is backup only (
expensive, poor quality etc.)?
Does it make sense to dedicate one link for outbound traffic and another
link for inbound traffic? | s*****g
发帖数: 1055 | 2 It does not make sense to dedicate one link for outbound and another link
for inbound, the reasons being, ISPs charge you based on your 95th
percentile
bandwidth usage inbound or outbound on top of your committed bandwidth, you
don't want to end up paying both providers for bandwidth outside your
commit, the other reason is unsymmetrical routing will confuse hell of you
when troubleshooting connectivity problems, it is something unavoidable but
you should try to avoid. | x*********n
发帖数: 28013 | 3 Usually most of enterprise will peer 2 different ISP to advertise their
routes.
1 as primary, the other one used as backup.
in most cases, my customers,ISP only advertised the default routes to
enterprise, because enterprise doesn't have to know too much.
outbound only affect on the edge router, so other iBGP doesn't care. | n**********l
发帖数: 271 | 4 Thanks. When is burstable billing usually used? Between network operators?
What I usually see appears to be just a fixed price for a CIR with not
burstable billing (unless I read it in the wrong way).
for bandwidth like this (HE)
Full 10 Gigabit Ethernet connection in any US, Canadian or European
Hurricane Electric POP for $0.75/Mbps with a one-year contract. ($7,500/
month, $2,500 setup fee, customer is responsible for all cross connect fees).
This one does not use burstable billing?
Does burstable billing offer any savings if billable use (95%) is lower than
CIR?
Why paying one provider overusage is preferred over paying both(paying ISP A
for 200mbps usage over paying ISP A and B each for 100mbps usage)?
Thanks
you
but
【在 s*****g 的大作中提到】
: It does not make sense to dedicate one link for outbound and another link
: for inbound, the reasons being, ISPs charge you based on your 95th
: percentile
: bandwidth usage inbound or outbound on top of your committed bandwidth, you
: don't want to end up paying both providers for bandwidth outside your
: commit, the other reason is unsymmetrical routing will confuse hell of you
: when troubleshooting connectivity problems, it is something unavoidable but
: you should try to avoid.
| n**********l
发帖数: 271 | 5 Thanks.
I was assuming peering at multiple locations so iBGP is involved.
If the customer only gets a default route, why use BGP? There are a lot of (
cheaper) ways to do outbound load balance via multiple WAN links.
For some reason my ISP has been pointing default at AS3549 (Level3) for the
past few months... All outbound traffic goes to AS3549 while inbound traffic
arrive via different peers (AS3549, AS6939 etc.) The first hop/last hop is
different from bidirectional traceroute which does confuse the hell out of
me.
(my IP replaced with fictional address of course)
Inbound
core1.fmt1.he.net> traceroute 8.1.2.3
Tracing the route to IP node (8.1.2.3) from 1 to 30 hops
1 <1 ms <1 ms <1 ms 184.105.213.66
2 71 ms 77 ms 71 ms 184.105.213.178
3 79 ms 79 ms 89 ms 209.51.161.114
4 85 ms 143 ms 97 ms 66.37.64.142
5 78 ms 77 ms 72 ms 69.77.128.114
6 72 ms 73 ms 78 ms 8.1.2.3
Outbound
tracert 216.218.252.161
Tracing route to core1.fmt1.he.net [216.218.252.161]
over a maximum of 30 hops:
1 <1 ms <1 ms <1 ms my-router
2 1 ms <1 ms <1 ms 8.1.2.1
3 5 ms 5 ms 5 ms 206.165.196.81
4 65 ms 75 ms 65 ms 64.209.105.42
5 72 ms 72 ms 73 ms 10gigabitethernet12-4.core1.pao1.he.net [72.
52.9
2.21]
6 74 ms 87 ms 75 ms core1.fmt1.he.net [216.218.252.161]
Trace complete.
Is this what they call routing at the edge? The edge I am connected to
directly peers with upstream?
【在 x*********n 的大作中提到】
: Usually most of enterprise will peer 2 different ISP to advertise their
: routes.
: 1 as primary, the other one used as backup.
: in most cases, my customers,ISP only advertised the default routes to
: enterprise, because enterprise doesn't have to know too much.
: outbound only affect on the edge router, so other iBGP doesn't care.
| s*****g
发帖数: 1055 | 6 That is why the industry uses 95th percentile method of billing, your bursty
traffic will mostly fall out side 95th percentile envelop, so you don't pay
, if you have too much bursty traffic you will need renegotiate with your
provider to have higher commit rate.
fees).
than
【在 n**********l 的大作中提到】
: Thanks. When is burstable billing usually used? Between network operators?
: What I usually see appears to be just a fixed price for a CIR with not
: burstable billing (unless I read it in the wrong way).
: for bandwidth like this (HE)
: Full 10 Gigabit Ethernet connection in any US, Canadian or European
: Hurricane Electric POP for $0.75/Mbps with a one-year contract. ($7,500/
: month, $2,500 setup fee, customer is responsible for all cross connect fees).
: This one does not use burstable billing?
: Does burstable billing offer any savings if billable use (95%) is lower than
: CIR?
| x*********n
发帖数: 28013 | 7 BGP or EBGP is the only protocol works with different AS.
so when you got BGP, you can advertise your internal network to public, most
time, when you are running ospf, it would be much easier for your ISP to
know which network we should advertise out.
on the other side, ISP may have some filter by as-path list or community-
list to filter out by accepting routes from customer.
the traffic goes different way is very normal, simply the ISP PE got
different ways to deal with traffic.
the edge router can be your edge router,(the router peer with ISProuter) or
the PE router just as its name provider's edge router.
I might be wrong, listen to saiwong. hehehe.
(
the
traffic
is
【在 n**********l 的大作中提到】
: Thanks.
: I was assuming peering at multiple locations so iBGP is involved.
: If the customer only gets a default route, why use BGP? There are a lot of (
: cheaper) ways to do outbound load balance via multiple WAN links.
: For some reason my ISP has been pointing default at AS3549 (Level3) for the
: past few months... All outbound traffic goes to AS3549 while inbound traffic
: arrive via different peers (AS3549, AS6939 etc.) The first hop/last hop is
: different from bidirectional traceroute which does confuse the hell out of
: me.
: (my IP replaced with fictional address of course)
| s*****g
发帖数: 1055 | 8 Please see inline
[saiwong] Not necessarily, if you don't have your own transport between
multiple locations, you don't need and can not run iBGP
(
[saiwong] You still want run BGP to have control of your own announcement.
the
traffic
is
[saiwong] I don't quite understand what you are saying, if you are multi-
homing to multiple ISPs, you must be importing the full Internet table, than
99.999% of your traffic won't hit the default route.
【在 n**********l 的大作中提到】
: Thanks.
: I was assuming peering at multiple locations so iBGP is involved.
: If the customer only gets a default route, why use BGP? There are a lot of (
: cheaper) ways to do outbound load balance via multiple WAN links.
: For some reason my ISP has been pointing default at AS3549 (Level3) for the
: past few months... All outbound traffic goes to AS3549 while inbound traffic
: arrive via different peers (AS3549, AS6939 etc.) The first hop/last hop is
: different from bidirectional traceroute which does confuse the hell out of
: me.
: (my IP replaced with fictional address of course)
| n**********l
发帖数: 271 | 9
[saiwong] I don't quite understand what you are saying, if you are multi-
homing to multiple ISPs, you must be importing the full Internet table, than
99.999% of your traffic won't hit the default route.
That's exactly my point. My ISP has multiple views of the full table but
elects to send all outbound traffic to one peer (AS3549). Inbound traffic
still arrives from different peers (at least my ISP does announce their
prefixes to all peers). This behavior kinda interferes with CDN because the
routes to CDN edge is suboptimal (at least for outbound). They can always
blame CDNs for that...
For example, CDN servers are deployed in one upstream AS2711. DNS resolution
sends me to the CDN edge in AS2711 but my outbound traffic all goes to
AS3549 (default route). There is a chance that my traffic goes across the
continent twice to reach a peer....
Even RFC1918 destinations are sent to AS3549 (my ISP loves default route so
much....)
The traceroute I posted in the above post is very confusing. I am one hop
away from upstream AS. The said they were doing routing at the edge. How can
this be done in a regional ISP (a couple counties)?
inbound
4 85 ms 143 ms 97 ms 66.37.64.142 (Upstream, AS16942)
5 78 ms 77 ms 72 ms 69.77.128.114 (my ISP)
6 72 ms 73 ms 78 ms 8.1.2.3 (me)
Outbound
1 1 ms <1 ms <1 ms 8.1.2.1 (my ISP gateway from DHCP)
2 5 ms 5 ms 5 ms 206.165.196.81 (my ISP)
3 65 ms 75 ms 65 ms 64.209.105.42 (Upstream, AS3459)
I noticed I always assume a transit provider environment instead of
enterprise....
【在 s*****g 的大作中提到】
: Please see inline
:
: [saiwong] Not necessarily, if you don't have your own transport between
: multiple locations, you don't need and can not run iBGP
: (
: [saiwong] You still want run BGP to have control of your own announcement.
: the
: traffic
: is
: [saiwong] I don't quite understand what you are saying, if you are multi-
| s*****g
发帖数: 1055 | 10 For outbound traffic, It could be that your ISP has settlement free peering
with Level3, yet it has to pay transit to other ASes, so it will always
prefer Level3 when delivering traffic from you, also note that Level(3)
originates 40% of Internet prefixes. Unfortunately you don't have much
control of this behavior if the destination prefix is not owned by you.
So what is your problem again? unsymmetrical routing is so normal.
than
the
resolution
【在 n**********l 的大作中提到】
:
: [saiwong] I don't quite understand what you are saying, if you are multi-
: homing to multiple ISPs, you must be importing the full Internet table, than
: 99.999% of your traffic won't hit the default route.
: That's exactly my point. My ISP has multiple views of the full table but
: elects to send all outbound traffic to one peer (AS3549). Inbound traffic
: still arrives from different peers (at least my ISP does announce their
: prefixes to all peers). This behavior kinda interferes with CDN because the
: routes to CDN edge is suboptimal (at least for outbound). They can always
: blame CDNs for that...
|
|
