[c-nsp] OSPF, load balancing and uneven links - design Q

[Netfortius]

Hi,

I have a multi-site environment, linked in a hub-and-spoke way, with the 
central router dual, triple or even quadruple T1 connected to each remote 
site. One of the T1s to each site carries also voice channels, thus leaving 
for data anywhere from 640 to 768 Kbps. 

In order to "protect" applications, in the past, we used to static route [map] 
traffic between remote LANs and core systems and the "hub" site (e.g. all 
traffic for application "n" (represented by the IPs of the main systems) will 
be carried over one of the full T1s, leaving the fractional one for other 
traffic, etc.)

We have recently implemented OSPF and QoS, in such a way that:
- critical application "n" is now identified by TCP port (consistent), and 
QoS-ed wit min bandwidth
- bandwidth-hog apps (also identified by ports, e.g. MS Exch traffic - TCP 
102, MS SMB & AD traffic - TCP 445 & ...) are being shaped in QoS

Doing the above we were hoping to diminish the amount of static routes, allow 
load balancing and proper failover (OSPF), and give everyone a share part of 
the network resources.

Problem: because we could not use the value of bandwidth for critical apps in 
the same amount over the full T1s as over the fractional ones, or the shaping 
value for bandwidth-hogs), we have chosen a "proportional" approach, instead, 
e.g.: critical apps getting 320 Kbps over the T1 will get only 
(640/1,500)*320 Kbps over the fractional T1s (I showed the math as it is more 
obvious how we did the calculation). Now - with this approach - we have 
problems when traffic of any sort takes the fractional route, and we 
obviously have inconsistent behavior. 

Note: we do "ip route-cache"

Questions: 
1. With the given data (T1s and frac T1s, and apps in need for QoS), what 
would be a better way to resolve the multi-link-uneven-bandwidth issue we are 
presented with (hopefully without revertin back to the old static routes and 
route maps, which would also make OSPF not needed, by add back to the admin 
issue a huge problem) - or is there a way to correct our attempts?
NOTE: I have looked into a MLPPP, but could not figure out how to make work 
with frac T1s and full T1s combined, and also not sure if this would be a 
good solution?!?
2. What is a good way to monitor the behavior of such traffic?
----------------
3. [OT] As we keep adjusting and re-adjusting the QoS (we are in the beginning 
of a process that needs a lot of tweaking), I am pulling the queues values 
and bandwidth, into MRTG. The problem is that every time we make a change, 
the OID changes, and we need to redo the configs - is there a more generic 
MIB/OIDs we could use?!?

TIA,
Stefan