rstorch1@netzero.net schreef:
> Please help me here. High availability and redundancy is not the
> same
> as load balancing, right? Load balancing can point to different
> servers that may very well be configured the same but unless those
> servers themselves are clustered or have HACMP or another product
> that
> makes them redundant or highly available, load balancing is
> technically does not mean the same thing as "high availability" or
> "redundant", right?
>
> Please help me find a way to explain this to management. They are
> stating that if you have a load balancer, you have redundancy.
>

rstorch1,


The point where the redundancy may fail is the load balancer itself. If
you do not make that component redundant, the load balancer will become
the single point of failure.

Behind the load balancer you will have redundancy, otherwise you cannot
balance the load.


Kind regards,


Jan Gerrit Kootstra

rstorch1@netzero.net wrote:
> Please help me here. High availability and redundancy is not the
> same
> as load balancing, right? Load balancing can point to different
> servers that may very well be configured the same but unless those
> servers themselves are clustered or have HACMP or another product
> that
> makes them redundant or highly available, load balancing is
> technically does not mean the same thing as "high availability" or
> "redundant", right?
>
> Please help me find a way to explain this to management. They are
> stating that if you have a load balancer, you have redundancy.
>

Cars have 4 tires, which balance the load. If one goes flat, you still have to
pull off the road to change.

Commercial trucks with two tires on the end of each axle have redundancy,
because if one tire fails, the other tire can carry the load until the next stop.

On 29 Oct, 00:36, rstor...@netzero.net wrote:
> Please help me here. High availability and redundancy is not the
> same
> as load balancing, right? Load balancing can point to different
> servers that may very well be configured the same but unless those
> servers themselves are clustered or have HACMP or another product
> that
> makes them redundant or highly available, load balancing is
> technically does not mean the same thing as "high availability" or
> "redundant", right?
>
> Please help me find a way to explain this to management. They are
> stating that if you have a load balancer, you have redundancy.

Yes, load balancing is nothing to do with redundancy, however you
would be hard put to find a load balancer which does not implement
some sort of failover on the nodes it balances.

C.

On Tue, 28 Oct 2008 17:36:12 -0700, rstorch1 wrote:

> Please help me here. High availability and redundancy is not the same
> as load balancing, right? Load balancing can point to different servers
> that may very well be configured the same but unless those servers
> themselves are clustered or have HACMP or another product that
> makes them redundant or highly available, load balancing is technically
> does not mean the same thing as "high availability" or "redundant",
> right?
>
> Please help me find a way to explain this to management. They are
> stating that if you have a load balancer, you have redundancy.

I've written a number of "technical recommendations" on such things as
work (covering HA, redundancy, and capacity as well). An easier way to
think about it is that (a) redundant components should be very far away
from each other, in order to recover in case of component/row/power/data
center failures. HA, on the other hand, should be very close to each
other - so that service levels can be maintained. (b) HA components
should be physically close to each other so that user/server access, as
well as server/disk access are as similar as possible (think of "a flat
latency profile" across the servers).

Bear in mind, above, you talk about load balancers. You're even
confusing "scaling for capacity" (what load balancers do -- even the name
implies it) and "scaling for redundancy". [This, I assume, is why we
have a 20 page internal paper on just this]. This is really probably the
most dangerous confusion of all -- since it opens your service to
degraded service levels.

HA and scaling for capacity are very different as well. HA aims to
ensure that there are a minimum of 2 components, where one can take over
for the other in case of failure. "Load balancers" typically serve a
purpose of distributing load across two systems when one cannot tolerate
the load. In HA (or redundancy, for that matter), you have to have
enough capacity to ensure that any component can replace another. With
load balancing, you're already making the assumption that a single system
cannot support the workload. In a LB system, if a (single) component
fails, you /will/ have degraded service.

In general:

Load Balancing is to scale capacity up. If a single component fails, you
*want* all components to fail together -- a "shared fate" design. This
allows your recovery mechanisms to cleanly kick in and restore your
service and service level. The overall utilization of the system as a
whole will not exceed 100% utilization, and components share the same
operational fate.

High Availability is to ensure seamless "fast" recovery of a single
system. It assumes that all components in the system are of equal
capacity and performance. The elements should be all in near proximity
to each other to ensure as "mirror-like" service and service level as
possible. The overall utilization of the system does not exceed 50%
utilization, and components are placed near to each other, but share
separate fates.

Redundancy involves multiple systems of equal capacity, but placed very
far apart, to ensure /availability/ of service during significantly
disruptive events. The overall utilization of the system does not exceed
50% utilization, the components are placed far apart, and share separate
fates.

While we may debate the semantics of "HA" vs. "Redundancy", as they're
defined above, they serve two different purposes. Confusing LB and
Availability, however, is a toxic combination. If you're using LB to try
and address Availability, you can see that you're getting neither. In LB
you want components to fail together, in Availability you want components
to fail separately. That prior sentence is the litmus test we use to
discern between them.

rstorch1@netzero.net wrote:
> Please help me here. High availability and redundancy is not the
> same
> as load balancing, right? Load balancing can point to different
> servers that may very well be configured the same but unless those
> servers themselves are clustered or have HACMP or another product
> that
> makes them redundant or highly available, load balancing is
> technically does not mean the same thing as "high availability" or
> "redundant", right?
>
> Please help me find a way to explain this to management. They are
> stating that if you have a load balancer, you have redundancy.

Obviously you want you balancing box to have backup.

Beyond that you have no useful redundancy if the boxes being balanced are
equal to the incoming queries. To have useful redundancy the boxes being
balanced have to run below capacity so on the failure of one or more the
remaining can provide the minimum acceptable response time.

--
I have a Covenant with God, a new book exposing Abraham's most
successful scam.
-- The Iron Webmaster, 4064
http://www.giwersworld.org/holo/ a8