At 8:57 PM +0000 10/1/02, Don wrote:
>I think it should be pointed out here that the land mine isn't even
>linking the customers IGP to the ISP's IGP.  The land mine is linking the
>IGP's of the customers to each other.  Consider having 1000 customers, all
>advertising their nets to this IGP, then consider what happens when one of
>them decides to makes a net entry that is flat out wrong.  For instance,
>customer A decides he now wants to add network B and starts advertising it
>in the IGP to the ISP.  Customer B and the ISP suddenly have major problems.
>The only way to prevent this to is install route filters for all 1000
>customers.  And fixing it after it happens is a major nightmare.  And do you
>really think every one of those 1000 customers will advertise their networks
>correctly?



>     Personally, I find it much easier and safer to do 1000 static routes.
>The solution below only works if the ISP has sole control of each of the
>1000 customers edge routers.  And again, 1000 static routes is more
>reasonable than adding the burden of managing 1000+ more routers.  Imagine
>the password nightmare (or did you really expect to put the same password on
>the routers of two different customers?).  So now we add a RADIUS server.
>     Don

Excellent points. I think many people assume static routes are 
unmanageable because they think they have to configure them manually. 
Not so.

I don't know of an ISP that doesn't use some automated tool, even a 
spreadsheet, to manage address space it assigns to customers. These 
things can print reports, which can become files! Assume all your 
customers get /24 blocks in 10.1.0.0, and you assign a /30 from 
10.0.0.0/23 to their link to you.

Example:
     LAN           WAN
10.1.0.0/24      10.0.0.0/30
10.1.1.0/24      10.0.0.4/30
10.1.2.0/24      10.0.0.8/30


Create a couple of conventions:
    -- the LAN router interface is the highest address in the subnet, in this
       case .254.
    -- the customer end of the WAN link is .1
    -- the ISP end of the WAN link is .2

Now, as you check off the address assignment, generate the statement:

   ip route  255.255.255.0 

and put it in a configuration library for the distribution router 
You can get this information into the distribution router without 
rebooting using a telnet script or various copy/merge operations into 
RAM or NVRAM.

In like manner, you can generate the default route for the remote 
router, and DNS RR records for everything.

>""Mike Bernico""  wrote in message
>[EMAIL PROTECTED]">news:[EMAIL PROTECTED]...
>>  I'm not sure I'm in complete agreement.  The network I work for has
>several
>>  distribution routers that contain around 1000 T1 speed customers.  If we
>>  were to static route each of their networks it would add about 1000 to
>1500
>>  lines of router configuration to the router.  That would definately add
to
>>  our maintenance and provisioning work and make troubleshooting harder on
>our
>>  techs.   While I agree statics are probably the most stable way, I'm not
>>  sure it's necessarily the best way to aggrigate high volumes of
customers.
>>  We currently use EIGRP at the edge with the stub command, OSPF or IS-IS
>>  would work just as well.  Regardless, we would never let our IGP, that
>>  extends to the CE router, touch their IGP.  About 98% of our customers
are
>>  not BGP customers though.
>>
>>  YMMV
>>  Mike
>>
>>  -------------------
>>  Mike Bernico                     [EMAIL PROTECTED]
>>  Illinois Century Network      http://www.illinois.net
>>  (217) 557-6555
>>
>>
>>  > -----Original Message-----
>>  > From: Howard C. Berkowitz [mailto:[EMAIL PROTECTED]]
>>  > Sent: Monday, September 30, 2002 11:37 AM
>>  > To: [EMAIL PROTECTED]
>>  > Subject: Re: OSPF for ISPs [7:54540]
>>  >
>>  >
>>  > At 2:58 PM +0000 9/30/02, Don wrote:
>>  > >Rather than run OSPF to customers, it is generally much
>>  > better to have
>>  > >them use a default route to the ISP and for the ISP to run
>>  > static routes to
>>  > >the customer.  OSPF to the customer is a huge land mine for
>  > > the ISP and
>>  > >should be avoided in almost every case.
>>  > >     Don
>>  >
>>  > I agree completely with Don that an ISP _never_ should link its IGP
>>  > to that of the customer.  Don't fall into the trap of assuming that
>>  > BGP needs a full routing table or will consume excessive resources.
>>  >
>>  > I remain confused why a default route wouldn't serve, unless there
>>  > are multiple connections between the ISP and customer. By "send the
>>  > block to the customer," do you mean the block is in the customer's
>>  > space?  You could certainly use a second static route, which can be
>>  > generated automatically as part of your address assignment (see my
>>  > NANOG presentation,
>>  > http://www.nanog.org/mtg-9811/ppt/berk/index.htm).
>>  >
>>  > If that's not appropriate, have the customer announce his two blocks
>>  > to you with BGP and receive default from your BGP.
>>  >
>>  > >
>>  > >
>>  > >""Chris Headings""  wrote in message
>>  > >[EMAIL PROTECTED]">news:[EMAIL PROTECTED]...
>>  > >>  Good morning all.  I was wondering if someone could lend
>>  > me a little help
>>  > >>  about engineering OSPF in the backbone for an ISP
>>  > network.  I just had a
>>  > >>  couple of questions and hopefully someone can give me
>>  > some guidance.or
>>  > >even
>>  > >>  some CCO links with some specific examples or better yet
>>  > any material
>>  > >>  anywhere.
>>  > >>
>>  > >>  Say, for example, that a customer has a small block of IP's and a
>>  > >>  distribution router knows where that block is, via a
>>  > connected route,
>>  > like
>>  > >a
>>  > >>  /30 on a serial link.  But later down the line the
>>  > customer requests an
>>  > >>  additional block of 64 IP addresses, what is the best way
>>  > to send this
>>  > >block
>>  > >>  to the customer?  Do I need to run OSPF on the customer
>>  > equipment?  If
>>  > the
>>  > >>  customer router is not running OSPF, how do the routers
>>  > know how to get
>>  > to
>>  > >>  this destination?  I assume via static routing???
>>  > >>
>>  > >>  Thanks as always.
>>  > >>
>>  > >  > Chris




Message Posted at:
http://www.groupstudy.com/form/read.php?f=7&i=54677&t=54540
--------------------------------------------------
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]

Reply via email to