BGP Restrictions

Discussion in 'Cisco' started by SP, Nov 7, 2003.

  1. SP

    SP Guest

    I have a customer who currently has a 2600 series router with a T1 to the
    Internet. They get their external IP space from a small local ISP. They
    have a small IP block (/29 mask). This customer wants to have redundant and
    load balancing Internet lines. I contacted their existing ISP to enquire if
    they supported BGP with the specifications that the customer AS no be used
    as a transit AS and that I wanted to receive default routes only.

    The ISP is stating that they cannot support BGP due to restrictions of the
    size of the IP block. Is a /29 too small for BGP? Is this a legitimate
    restriction of the ISP?
    SP, Nov 7, 2003
    1. Advertisements

  2. SP

    Ivan Ostres Guest

    Technically, no. Practicaly Yes. It's kind of a hard to get ASN for such a
    small block...

    Ivan Ostres, Nov 7, 2003
    1. Advertisements

  3. Generally the smallest prefix size that is accepted in the global BGP
    routing tables is a /24.

    The desire to multihome, however, is usually sufficient justification
    to get a /24 from an ISP, and is always sufficient justification to
    get an AS from ARIN. (I assume other RIRs have the same policy on

    Terry Baranski, Nov 8, 2003
  4. SP

    Andre Beck Guest

    If it comes from the swamp or other PI allocation space. From well known
    RIR blocks that are typically allocated as /19s, /20s or /21s, it seems
    to be common practice that only announcements matching these allocation
    sizes are accepted, to prevent those who try to load balance by announcing
    their allocation in parts from sucking up others resources.
    ASN-wise yes. But just the desire to multihome doesn't justify a /24 in
    RIPE area. A /24 (or shorter) is assigned only if there is well documented
    need for that amount of addresses (and "I need them for no other reason
    but to get global routing" doesn't count).
    Andre Beck, Nov 12, 2003
  5. There's still some question as to just how common that is. Verio is
    the only Tier-1 I know of that does it. Routing table growth seems to
    be less of a concern these days relative to a few years ago when it
    was exponential.
    I was referring to PA space. Are you saying that a multihomer can't
    get a PA /24 from an ISP who gets its IP space from RIPE?

    Terry Baranski, Nov 14, 2003
  6. Not exponential... Wrong word. I think the word I was looking for was
    "a lot higher than it is now." 2000 was a bad year for routing table
    growth but the rate has decreased substantially since then. This,
    along with cheaper memory prices and faster hardware overall, may
    cause providers like Verio to rethink their BGP filtering strategies.

    Terry Baranski, Nov 14, 2003
  7. SP

    Andre Beck Guest

    There was a very popular Cisco platform (the 720x with NPE200) that would
    still serve a number of ISPs well if it just could deal with todays DFZ.
    But you cannot plug more than 128MiB into these blades and so you cannot
    plug them into the DFZ with CEF enabled and full tables. Luckily folks
    found out the NPE225 could eat more SDRAM than was initially specified...
    Not if he has no other reasonable justification for beeing assigned a /24
    and the ISP refuses to support a lie. This is independent of PA or PI.
    The potential multihomer could also try to become a LIR and just use
    part of the allocation he gets for his own assignment. The RIPE, however,
    requires an initial need for 25% of the space and a projected need of 50%
    IIRC - so no, you cannot become a LIR and get a /21 to announce and then
    just use one /27 out of that, either. That is harsh, but it is the price
    of conservation, especially outside the US. And it's rather more politics
    than tech. Beeing able to pace the growth of the global table by making
    it harder to become BGP multihomed for non-ISPs with very small address
    ranges seems to be an accepted side effect of the conversation rules.
    Andre Beck, Nov 15, 2003
  8. SP

    Andre Beck Guest

    Well, at the climax of .com, it really looked like it would grew expo-
    nential. Until the bubble blew up...
    When .com turned into .gone, the table took a hickup and settled for
    a while. Since late 2002, it seems to be constantly growing again, and
    speed slowly increases again, too. It's just about to exceed 128Ki
    routes (bye bye 128Ki CAMs)...
    IMO this never was a problem of memory prices. It was a problem of
    having to replace a complete chassis just to upgrade to some silly
    256MiB. For gods sake some kind soul found out the NPE225 (which is
    the best NPE that still runs in a non-VXR 720x) can actually digest
    a straight CL2 256MiB DIMM. With .gone still depressing on us and the
    overcapacity deployed before that, there is likely some pause in the
    swapout of old equipment. So at the moment, I don't really expect
    filter policies to change soon.
    Andre Beck, Nov 15, 2003
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.