[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: why i should like pibs
The Industry Realities COPS-PR & its PIBs Address:
Dynamic Edge vs. Static Core -
* COPS exists to address the Dynamic Provisioning needs of the Network's
Edge.
- Access Control Requests (from a variety of signaled protocols)
- Controlled allocation of network resources
- Micro-flow to aggregate traffic characterization
- QoS
- Security
- Tunneling
- Usage feedback
* Meanwhile the Network Core tends to be static and deal with traffic
aggregates being mainly concerned with routing & moving packets as fast as
possible with minimal overhead.
... And ... It's a Multi-protocol World, COPS-PR is the Integrator. It
closes the loop by tying together diverse signaling and provisioning
mechanisms. Access control/session initiation/QoS requests come in and
dynamically allocating/provisioning device resources based on policy goes
out. Finally policy usage feedback closes the loop.
Qualitative:
* Enables entirely new classes of dynamic and integrated services that would
not be possible without it.
* It can integrate resource allocation and control for pretty much every
in-band + out-of-band signaling protocol AND pretty much any form of
resource Provisioning (eg. DiffServ).
* BIG and RELIABLE TRANSACTIONS with rollback, failover and synchronization
built-in.
* It just works. It pulls outsourcing and provisioning together in one nice
state-driven solution. Stateful means it maintains the state of all sessions
on a device at all times as well as the resources they consume.
* Completely event driven.
* Device capabilities reporting is integral to the PIBs. Everything the
device can syntactically parse and semantically do is precisely yet
generically reported to the PDP.
* Implementations are easy... COPS-PR sends you all your data in one
complete reliable transaction. SNMP can throw all your attributes into a
blender, so your implementation needs to be able to unscramble what comes
out the other side, loss and all.
* Three levels of easily understandable security.
* New and improved data model and definition language with a consistent
theme of a data-flow throughout. Enables better and improved tools over SNMP
to make the job of implementers and users even easier!
* Solves the multi-manager problem, data instances cannot overlap managers
in COPS-PR and, thus, managers cannot step on one-another's toes.
* NO row-status, owner description strings, storage-types, etc. to deal with
AT ALL... And good riddance.
Quantitative:
* It can do one RTT provisioning based on outsourced events = well within
call-setup time = as fast as is possible between two remote systems.
* Intrinsically 10x more efficient on the wire than SNMP (1/10th the data to
xfer) for e.g. the ever common DiffServ IP-filter tables. Efficiency
multiplies with the more attributes you have in a row.
* Faster, better & more. Change 10000 DiffServ Filter+meter+action entries
through a T1 line with a 10msec RTT for 48byte packets:
SNMP=((10000*8*498)/1540000)+((2*10000)/100)=226 Seconds.
COPS=((10000*8*(498/10))/1540000)=2.59 Seconds.
Is 100x improvement sufficiently better? And the multiple goes up with the
more data you xfer. ... adding bandwidth doesn't help, it's that dang RTT.
-Dave
> -----Original Message-----
> From: Randy Bush [mailto:randy@psg.com]
> Sent: Monday, March 18, 2002 6:13 AM
> To: rap@ops.ietf.org; diffserv@ietf.org
> Cc: ipsec-policy@vpnc.org
> Subject: why i should like pibs
>
> wearing my iesg hat but being just a stupid operator, i am trying to
> understand the pib/mib controversy. fyi, i currently use snmp heavily
> for monitoring devices on my network. i configure using large db-driven
> code and spew text-based cli to the devices.
>
> let's assume i want to take the leap to a binary, as opposed to textual,
> configuration language. i.e. for some reason(s) [which we will PLEASE
> NOT discuss here] i decide to move from pushing text-based cli configs
> out to pushing a binary format.
>
> hence, i would have to push my vendors to implement snmp/cops writes for
> all configuration aspects of all devices. this would be big cost for
> both me and for my vendors.
>
> why would cops/pibs be significantly better (remember it has to replace
> my current investment, so it can not be 'just as good') than snmp/mibs?
>
> randy