Speaking as WG member,
This summary really helped clarify things for me. Calculating paths
based on a power metric seems like a valid stand-alone use case, which
seems like a worthwhile thing for the WG to work on.
Thanks,
Chris.
(As wg member)
"Bonica, Ron" <[email protected]> writes:
Aijun, GenChen, Benoit, JinMing
Each of you have expressed concern regarding a distributed approach to the Power
Conserving Path Placement Strategy (PCPPS). The following is a response.
A Power Conservation Strategy includes many components. First among them is the
Path Placement Component. The Path Placement Component concentrates traffic onto
relatively few network resources during periods of low demand and redistributes
traffic during periods of high demand. It *does not* power interfaces up and
down. Its only function is path computation. When it computes paths, it
optimizes for energy efficiency, leaving some network resources idle or nearly
idle.
Another component is the Sleep Management Component. The Sleep Management
Component is configured with criteria that must be satisfied before a network
resource is powered down. It monitors network resources and powers them down
when they satisfy the configured criteria. It also powers them up again when the
power up criteria are satisfied.
The Power Placement Component and Sleep Management Component are decoupled from
one another. In the simplest case, one could run a Path Placement Component
without a Sleep Management Component. (Although I can't imagine why). One could
also run a Sleep Management Component without a Path Placement Component. In
some networks, and for some power down criteria, this makes sense.
In a more complex case, the Sleep Management component may require information
from the Path Placement component to determine if an interface satisfies the
power-up/power-down criteria. For example, the Sleep Management algorithm may
prevent interfaces that support LSPs from being powered down.
Draft-many-teas-power-steering and draft-many-lsr-power-groups describe the Path
Placement Component. They *do not* describe the Sleep Management function. This
is beyond their scope.
Many of your comments address the Sleep Management function. For example:
* You will not find an instruction to power an interface up or down in either of
the current drafts
* The problem of oscillation due to frequent power-up/power-down events is
beyond the scope of the current documents, because they do not power network
resources up or down.
Finally, regarding the question of distributed versus centralized
implementation.....
Path Placement Strategies rely on CSPF and the CSPF relies on the LSDB/TED. Each
node must maintain an identical LSDB copy.
In the past, we have populated the LSDB/TED in a distributed manner, using an
IGP, or in a centralized manner, using a PCE and BGP-LS. Both approaches work.
Some customers prefer one while other customers prefer the other. The Power
Conserving Path Placement Strategy is no different from legacy path placement
strategies. There is no reason why it should be limited to one approach or the
other.
Ron
_______________________________________________
Lsr mailing list -- [email protected]
To unsubscribe send an email to [email protected]
