Hi Naiming, et al,
On Dec 29, 2006, at 12:32 AM, Naiming Shen wrote:
Curtis,
Curtis Villamizar said the following on 12/28/2006 05:58 PM:
In message <[EMAIL PROTECTED]>
Naiming Shen writes:
Curtis Villamizar said the following on 12/28/2006 03:03 PM:
In message <[EMAIL PROTECTED]>
Acee Lindem writes:
I don't think the problems above warrant allocating an OSPFv3
prefix
option bit since most of them should be solved outside of
OSPFv3. For
example, I'd hope modern routers wouldn't default to per-packet
ECMP
forwarding.
I don't think any router in use in a provider has done per-
packet ECMP
in a long time but there is no telling what an off brand enterprise
sized router will do. I recently encountered a DSL router that
had to
But this is not a problem with OSPF specifically. For dynamic
routing
information on prefixes(including anycast route), one can otionally
add a sub-tlv/attribute/option-bit/opaque-option to signal a
"Do-Not-Per-Packet-ECMP" status of the route for BGP/ISIS/OSPF/etc,
and it's up to the implememtation to automatically disable the
per-packet loadsharing even it's explicitly configured on outbound
interfaces. Anycast route maybe only one of the applications
which can benefit with this trick.
- Naiming
Naiming,
Just about everything benefits from this trick. If you do per packet
ECMP and the two legs have unequal delay you get out of order
delivery. Real time applications are not at all fond of getting out
of order delivery. TCP can have false triggers of the "fast restart"
mechanism which dramatically slow down TCP transfers. Per-packet
ECMP
has been proven in the field to be a very bad idea and the reasons it
causes problems are very well understood.
Per-packet ECMP should not be done under any circumstances. Hash
based load split is circa 1986/1987 technology and works. Use it.
See RFC2991 and RFC2992 for more detailed explanation.
It is unfortunate that routers still allow per-packet ECMP to be
configured. At least providers know not to use this feature. But we
can all be sure that somewhere out there someone ignorant of the
problems has enabled it.
I completely agree. Per-packet ECMP should not be used normally,
especially in provider's networks. But I did experience some cases
( a while ago ) on international links, there were some dominant
prefixes which led to very uneven link bandwidth usages due to flow
based algorithm. I think if folks in the routing area still think
things like anycast having this issue, then to signal this
attribute is
useful and can be easily done; otherwise, there is absolutely no issue
in terms of anycast prefiexes(ipv4 or ipv6).
Curtis
ps - Maybe the attribute semantics should be an "OK to per packet
load
split traffic to this route" and all we have to do is never use
it. :-)
Well, I don't think any one wants to set this flag in the first
place;-)
So, with the limited OSPFv3 prefix bits I don't see enough motivation to
allocate one for the purpose of identifying anycast addresses. What
we've
discussed is a hypothetical problem with anycast and ECMP and not one
anyone is complaining about.
Thanks,
Acee
- Naiming
be reboot to make any configuration change effective so bad design
even today is clearly possible.
Curtis
_______________________________________________
OSPF mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/ospf
