BGP question - 未名空间MITBBS历史存档

BGP question# EmergingNetworking - 热门网络技术

j*r2006-03-09 08:03

1 楼

All Cisco experts,
We are running bgp on many routers for Internet load balance. We have two T1
circuits from one ISP, let's focus on these two T1 circuits.
On our side, the two Ts terminate on the same router, on ISP side, they
terminate on different routers. Most of time, one circuit got 100% saturated,
the other one has only 20% usage.
Is there any way easy to adjust the load between the circuits? we only care
about the inbound traffic.

b*e2006-03-09 08:03

2 楼

For NAT, I was thinking that you NAT half of your address to one next hop, and
the other half to the other next hop and you don't advertise your /24 block
to SP at all.

【在 j****r 的大作中提到】

: All Cisco experts,
: We are running bgp on many routers for Internet load balance. We have two T1
: circuits from one ISP, let's focus on these two T1 circuits.
: On our side, the two Ts terminate on the same router, on ISP side, they
: terminate on different routers. Most of time, one circuit got 100% saturated,
: the other one has only 20% usage.
: Is there any way easy to adjust the load between the circuits? we only care
: about the inbound traffic.

j*r2006-03-09 08:03

3 楼

We can't do that.
Thanks for the reply, anyway.

and

【在 b******e 的大作中提到】

: For NAT, I was thinking that you NAT half of your address to one next hop, and
: the other half to the other next hop and you don't advertise your /24 block
: to SP at all.

z*r2006-03-09 08:03

4 楼

didn't see it, but anyway, if that's the case, MED won't work.
Let's think about this issue in another way before we jump to other approaches
.
You have 2 T1's connected to 2 ISP routers, why? I guess the reason is for 1)
redundancy, 2)load sharing, so, what model is your router? and what models are
the ISP edge routers? if their routers are higher end, maybe you can think
about your ISP's advice, just move the 2 T1's to one sigle router. This makes
things much easier. for you, you don't need t

【在 j****r 的大作中提到】

: We can't do that.
: Thanks for the reply, anyway.
:
: and

j*r2006-03-09 08:03

5 楼

Your assumption is correct. Anything not clear, you can use "assumption",
it should work in my situation, hehe.
Our ISP says they do support MED, but it won't work, the only solution
is to move two circuits to the same router.
Cisco says we have to use OER.
Any ideas? thanks.

【在 z**r 的大作中提到】

: didn't see it, but anyway, if that's the case, MED won't work.
: Let's think about this issue in another way before we jump to other approaches
: .
: You have 2 T1's connected to 2 ISP routers, why? I guess the reason is for 1)
: redundancy, 2)load sharing, so, what model is your router? and what models are
: the ISP edge routers? if their routers are higher end, maybe you can think
: about your ISP's advice, just move the 2 T1's to one sigle router. This makes
: things much easier. for you, you don't need t

z*r2006-03-09 08:03

6 楼

didn't see it, but anyway, if that's the case, MED won't work.
Let's think about this issue in another way before we jump to other approaches
.
You have 2 T1's connected to 2 ISP routers, why? I guess the reason is for 1)
redundancy, 2)load sharing, so, what model is your router? and what models are
the ISP edge routers? if their routers are higher end, maybe you can think
about your ISP's advice, just move the 2 T1's to one sigle router. This makes
things much easier. for you, you don't need t

【在 j****r 的大作中提到】

: Your assumption is correct. Anything not clear, you can use "assumption",
: it should work in my situation, hehe.
: Our ISP says they do support MED, but it won't work, the only solution
: is to move two circuits to the same router.
: Cisco says we have to use OER.
: Any ideas? thanks.

z*r2006-03-09 08:03

7 楼

waste of time, if he moves the 2 T1's together, the bgp auto load balancing on
the ISP side will take of everything, no use of MLPPP
still waste of time, he has auto load balancing turned on right now, but this
takes care of only the outbound traffic, which is not op's concern

j*r2006-03-09 08:03

8 楼

We can't do that.
Thanks for the reply, anyway.

and

【在 b******e 的大作中提到】

: For NAT, I was thinking that you NAT half of your address to one next hop, and
: the other half to the other next hop and you don't advertise your /24 block
: to SP at all.

b*e2006-03-09 08:03

9 楼

For NAT, I was thinking that you NAT half of your address to one next hop, and
the other half to the other next hop and you don't advertise your /24 block
to SP at all.

【在 j****r 的大作中提到】

: We can't do that.
: Thanks for the reply, anyway.
:
: and

j*r2006-03-09 08:03

10 楼

This is not the only bgp router in our network, we have other routers with
bgp sessions to different ISPs.
We have made decision last week to move the two Ts together and use ppp
multilink. I was just try to find out a way to avoid this time-consuming
change if possible.
Thanks anyway to everybody for the ideas.

approaches
)
are
makes

【在 z**r 的大作中提到】

: didn't see it, but anyway, if that's the case, MED won't work.
: Let's think about this issue in another way before we jump to other approaches
: .
: You have 2 T1's connected to 2 ISP routers, why? I guess the reason is for 1)
: redundancy, 2)load sharing, so, what model is your router? and what models are
: the ISP edge routers? if their routers are higher end, maybe you can think
: about your ISP's advice, just move the 2 T1's to one sigle router. This makes
: things much easier. for you, you don't need t

z*r2006-03-09 08:03

11 楼

waste of time, if he moves the 2 T1's together, the bgp auto load balancing on
the ISP side will take of everything, no use of MLPPP
still waste of time, he has auto load balancing turned on right now, but this
takes care of only the outbound traffic, which is not op's concern

z*r2006-03-09 08:03

12 楼

first of all, it's nice to see folks to seperate the inbound and the outbound
traffic when considering multihoming.
there are dozens of multihoming scenarios, and your situation fits into the
one as you described exactly. however, there are a couple of things you have
not mentioned yet, are you learning the default routes only? are you learning
a default route from one link and full or partial full bgp routes from another
link? you don't disable any load balancing (which is on by default) manua

z*r2006-03-09 08:03

13 楼

inbound
I know, the reason I was talking about the whole bgp table was, op mentioned
cisco OER, I think OER is something to deal with outbound traffic, as well as
some limitation of capacity. I have never been saying outbound solution.
care
I considered AS-PATH in the beginning actually, however, I think AS-PATH is
more useful when we have multihome to multiple AS's (multiple ISPs), the best
inbound control in op's situation is still MED I believe. :)
network

c*a2006-03-09 08:03

14 楼

OP was asking about inbound traffic, how can whole BGP table help his inbound
traffic?
1> can we prepend AS on the saturated links
2> dynamically change AS using a script based on load of links, but more care
needs to be taken to avoid flip flops, a delay may introduced
3> introduce global load balance via DNS (say set two IP's inside your network
, polling from external, ie, ISP, and dynamically change your IP?)
But if all solutions work around are more expensive than simply upgade the
bandwidt

【在 z**r 的大作中提到】

:
: inbound
: I know, the reason I was talking about the whole bgp table was, op mentioned
: cisco OER, I think OER is something to deal with outbound traffic, as well as
: some limitation of capacity. I have never been saying outbound solution.
: care
: I considered AS-PATH in the beginning actually, however, I think AS-PATH is
: more useful when we have multihome to multiple AS's (multiple ISPs), the best
: inbound control in op's situation is still MED I believe. :)
: network

z*r2006-03-09 08:03

15 楼

didn't see it, but anyway, if that's the case, MED won't work.
Let's think about this issue in another way before we jump to other approaches
.
You have 2 T1's connected to 2 ISP routers, why? I guess the reason is for 1)
redundancy, 2)load sharing, so, what model is your router? and what models are
the ISP edge routers? if their routers are higher end, maybe you can think
about your ISP's advice, just move the 2 T1's to one sigle router. This makes
things much easier. for you, you don't need t

【在 j****r 的大作中提到】

: This is not the only bgp router in our network, we have other routers with
: bgp sessions to different ISPs.
: We have made decision last week to move the two Ts together and use ppp
: multilink. I was just try to find out a way to avoid this time-consuming
: change if possible.
: Thanks anyway to everybody for the ideas.
:
: approaches
: )
: are

c*a2006-03-09 08:03

16 楼

hoho, seems to be same as my suggestion

【在 j****r 的大作中提到】

: This is not the only bgp router in our network, we have other routers with
: bgp sessions to different ISPs.
: We have made decision last week to move the two Ts together and use ppp
: multilink. I was just try to find out a way to avoid this time-consuming
: change if possible.
: Thanks anyway to everybody for the ideas.
:
: approaches
: )
: are

j*r2006-03-09 08:03

17 楼

Thanks for the sample, it doesn't work for us.
The problem is, we only have one /24 network, we cannot advertise two
different networks with different MEDs to the two BGP peers, the ISP
won't take /25.

adjusting

【在 z**r 的大作中提到】

: didn't see it, but anyway, if that's the case, MED won't work.
: Let's think about this issue in another way before we jump to other approaches
: .
: You have 2 T1's connected to 2 ISP routers, why? I guess the reason is for 1)
: redundancy, 2)load sharing, so, what model is your router? and what models are
: the ISP edge routers? if their routers are higher end, maybe you can think
: about your ISP's advice, just move the 2 T1's to one sigle router. This makes
: things much easier. for you, you don't need t

j*r2006-03-09 08:03

18 楼

We can't do that.
Thanks for the reply, anyway.

and

【在 b******e 的大作中提到】

: For NAT, I was thinking that you NAT half of your address to one next hop, and
: the other half to the other next hop and you don't advertise your /24 block
: to SP at all.

z*r2006-03-09 08:03

19 楼

what? ft, then it's gonna be a totally different story. I thought you can
advertise multiple routes to your ISP. you sure they won't take /25?
if they won't, then I don't think BGP can solve this issue. let's figure out
other ways. let's dig into cisco OER again then, hehe

【在 j****r 的大作中提到】

: Thanks for the sample, it doesn't work for us.
: The problem is, we only have one /24 network, we cannot advertise two
: different networks with different MEDs to the two BGP peers, the ISP
: won't take /25.
:
: adjusting

m*t2006-03-09 08:03

20 楼

Will VRRP help to load sharing?

【在 j****r 的大作中提到】

: We can't do that.
: Thanks for the reply, anyway.
:
: and

j*r2006-03-09 08:03

21 楼

Your assumption is correct. Anything not clear, you can use "assumption",
it should work in my situation, hehe.
Our ISP says they do support MED, but it won't work, the only solution
is to move two circuits to the same router.
Cisco says we have to use OER.
Any ideas? thanks.

【在 z**r 的大作中提到】

:
: what? ft, then it's gonna be a totally different story. I thought you can
: advertise multiple routes to your ISP. you sure they won't take /25?
: if they won't, then I don't think BGP can solve this issue. let's figure out
: other ways. let's dig into cisco OER again then, hehe

z*r2006-03-09 08:03

22 楼

waste of time, if he moves the 2 T1's together, the bgp auto load balancing on
the ISP side will take of everything, no use of MLPPP
still waste of time, he has auto load balancing turned on right now, but this
takes care of only the outbound traffic, which is not op's concern

【在 c*a 的大作中提到】

:
: hoho, seems to be same as my suggestion

j*r2006-03-09 08:03

23 楼

We have been thinking about MED since the very beginning, our ISP and Cisco
both said it wouldn't work. Personally I still think that's the best easiest
way to do it, just need to get more details.

as
best

【在 z**r 的大作中提到】

:
: waste of time, if he moves the 2 T1's together, the bgp auto load balancing on
: the ISP side will take of everything, no use of MLPPP
: still waste of time, he has auto load balancing turned on right now, but this
: takes care of only the outbound traffic, which is not op's concern

z*r2006-03-09 08:03

24 楼

first of all, it's nice to see folks to seperate the inbound and the outbound
traffic when considering multihoming.
there are dozens of multihoming scenarios, and your situation fits into the
one as you described exactly. however, there are a couple of things you have
not mentioned yet, are you learning the default routes only? are you learning
a default route from one link and full or partial full bgp routes from another
link? you don't disable any load balancing (which is on by default) manua

j*r2006-03-09 08:03

25 楼

This is not the only bgp router in our network, we have other routers with
bgp sessions to different ISPs.
We have made decision last week to move the two Ts together and use ppp
multilink. I was just try to find out a way to avoid this time-consuming
change if possible.
Thanks anyway to everybody for the ideas.

approaches
)
are
makes

【在 z**r 的大作中提到】

: didn't see it, but anyway, if that's the case, MED won't work.
: Let's think about this issue in another way before we jump to other approaches
: .
: You have 2 T1's connected to 2 ISP routers, why? I guess the reason is for 1)
: redundancy, 2)load sharing, so, what model is your router? and what models are
: the ISP edge routers? if their routers are higher end, maybe you can think
: about your ISP's advice, just move the 2 T1's to one sigle router. This makes
: things much easier. for you, you don't need t

c*a2006-03-09 08:03

26 楼

OP was asking about inbound traffic, how can whole BGP table help his inbound
traffic?
1> can we prepend AS on the saturated links
2> dynamically change AS using a script based on load of links, but more care
needs to be taken to avoid flip flops, a delay may introduced
3> introduce global load balance via DNS (say set two IP's inside your network
, polling from external, ie, ISP, and dynamically change your IP?)
But if all solutions work around are more expensive than simply upgade the
bandwidt

【在 z**r 的大作中提到】

: first of all, it's nice to see folks to seperate the inbound and the outbound
: traffic when considering multihoming.
: there are dozens of multihoming scenarios, and your situation fits into the
: one as you described exactly. however, there are a couple of things you have
: not mentioned yet, are you learning the default routes only? are you learning
: a default route from one link and full or partial full bgp routes from another
: link? you don't disable any load balancing (which is on by default) manua

z*r2006-03-09 08:03

27 楼

I think NAT is not a choice for him, since he even has the whole BGP routes (
not only 2 defaut routes)

and

【在 b******e 的大作中提到】

: For NAT, I was thinking that you NAT half of your address to one next hop, and
: the other half to the other next hop and you don't advertise your /24 block
: to SP at all.

z*r2006-03-09 08:03

28 楼

which
^^^^^^^^^^^^^^^^^^^
I mean from, hehe

【在 z**r 的大作中提到】

: I think NAT is not a choice for him, since he even has the whole BGP routes (
: not only 2 defaut routes)
:
: and

j*r2006-03-09 08:03

29 楼

depends
originate
We knew ISP could set local-pref based on source IPs or ASes. The problem
is, ISP might do things once, but won't do adjustment for us daily. BGP
adjustment is a long term daily work.
I am still thinking about this. The ISP is not taking anything longer than
/24, I am not sure if we can do it based on source IPs or ASes.
to
The ISP is not taking anything longer than /24.
滑阴道减少不适感，但在阴茎头部不要涂得太多，否则容易使避孕套脱落。
(4)射精后不要将阴茎长时间留在阴道内，应在阴茎未软缩之前，用手按住套
口使阴茎连同避孕套一起从阴道内抽出，以防阴茎软缩后避孕套脱落在阴道内
或精

【在 b******e 的大作中提到】

: For NAT, I was thinking that you NAT half of your address to one next hop, and
: the other half to the other next hop and you don't advertise your /24 block
: to SP at all.

c*a2006-03-09 08:03

30 楼

One req from OP (dynamically L/B) is almost impossible with BGP without doing
more intelligent monitoring/dynamic configuration
reading some white paper now about L/B, quite complicated, and this is even
only with static (predefined algorithm) L/B