On Tue, March 23, 2010 7:24 pm, Yoav Nir wrote:
>
> On Mar 23, 2010, at 6:05 PM, Dan Harkins wrote:
>
>>
>> Hi,
>>
>> "hot standby" implies a box sitting ("hot") twiddling its thumbs doing
>> little but waiting for another box to fail ("standby"). It's the VRRP
>> model.
>
> And that's exactly what I want to describe. Well, not twiddling its
> thumbs. The standby is synchronizing state with the active member, but
> it's not doing any IKE or IPsec
Well don't use such a limiting term to describe a behavior that is
not so limited. Not all HA is that small subset you want to describe.
>> There is a HA model which supports dynamic load balancing as well as
>> active session failover. Nodes in such a cluster are not "standby". They
>> each have loads that they can shed and add to based upon some heuristic.
>> A neat attribute of such a system is that an IPsec SA can be established
>> on node A, move to node B after a while, and come back to A some time
>> later without any actual node failure. State moves around to keep the
>> cluster balanced.
>
> Failure is just used as an example of why a certain SA failed over to
> another member. It is by no means the only reason. Still, what you are
> describing is a model that provides both high-availability and load
> balancing, and that is the reason we're moving away from calling the first
> model "high availability".
Of course it's not the only reason. But you're missing the point. The
point is that the reason doesn't matter! You want to describe a particular
reason-- the "master" crashed and all state went over to the "hot standby"--
not the generic concept of state moving.
So don't call it "high availability" then. But "hot standby" is worse.
>> I would very much prefer "session failover" to "hot standby" and a
>> mild preference of "load balancing" over "load sharing". An HA model
>> doing VRRP could be termed "session failover" but the HA model described
>> above really can't be called "hot standby". And load can be shared but
>> just sharing a load can result in a mis-balanced cluster if sessions on
>> one node terminate naturally and it sits doing little while another node
>> whose sessions haven't terminated is huffing-and-puffing. Balancing can
>> imply sharing but not vice versa.
>
> "Session failover" sounds to me more like a description of an event than a
> type of cluster.
So what? Are you suggesting that a type of cluster that is not a
"hot standby" is not worthy of terminology in IPsecME?
Your term is severely limiting. I like "session failover". If you don't
then come up with a term that generically describes HA and not the
particular style of HA that you are accustomed to.
Dan.
_______________________________________________
IPsec mailing list
IPsec@ietf.org
https://www.ietf.org/mailman/listinfo/ipsec
