OER-Load Balancing

hclisschennai · ‎05-18-2009

Hi Everybody,

I am configuring OER in 1841 Router. I have 2 ISPs

ISP1:4 Mbps

ISP2:8 Mbps

I have configured

" max-range-utilization percent 15"

under master controller and two default routes, with one default route having higher metric than other.

ip route 0.0.0.0 0.0.0.0 ISP1

ip route 0.0.0.0 0.0.0.0 ISP2 220

My understanding is that when ISP1 link reaches 1.2 Mbps (15% of 8 Mbps), the traffic some start flowing in ISP2 also along with ISP1.

Is this true. Did I understood in the right way?

If this is the case how the traffic is distributed between two ISP links? whether it per-packet Loadbalancing or Per-destination loadbalancing?

Can you please guide me in this concepts

Thanks in advance

RBK

Joseph W. Doherty · ‎05-18-2009

Max-range-utilization would be the allowable load balance difference between links, as load percentages per path (I believe). If the multiple paths have a load imbalance more than what they're configured for, OER will attempt to bring the links back into balance.

If one link was at x% load, the other should be +/- max-range-utilization percent. I.e. the actual bandwidths would depend on the links' bandwidths. [edit - in other words, if the two link load percentages differ by more than 15%, in your example, OER would attempt to better balance them.]

It's unclear why you have two default routes with different metrics. Is there a reason why having both in the route table to begin with is unacceptable? [edit - perhaps you're thinking because the one link offers 2x the bandwidth of the other. If there's a large difference, that might make sense, but unsure 4 vs. 8 Mbps should be so treated. OER has some delay to its flow management and flows directed to a slower link would be limited by the link's bandwidth regardless of OER balancing.]

Outbound traffic is distributed by OER injecting new destination routes (or by using PBR) for active traffic flows to force some through just one path. There's no, per se, per-packet or per-destination load balancing, there's link load balancing by directing flows. (NB: depending on OER configuration, OER could direct flows by flow performance over just link balancing. I.e. OER can intentionally imbalance links to provide the best individual flow performance.)

hclisschennai · ‎05-19-2009

Hi Joseph,

Thanks for your comments. But i find difficult to understand what you are trying to convey.

I will put it in simple ways. I have one Internet proxy server for outbound Internet connection with IP 218.248.6.2

I have 2 Internet ISP links A & B with A=4 Mbps and B=8 Mbps

When "max-range-utilization percent 15" under oer master what will happen with the traffic flows.

ip route 0.0.0.0 0.0.0.0 ISPB

ip route 0.0.0.0 0.0.0.0 ISPA

My understanding is Initiall traffic will flow through ISP B and if ISP B link utilization reaches 1.2 Mbps (15% of 8 Mbps), some traffic will flow through ISP A.

After this situation, how the traffic path is choosen. Whether forth coming flows are taken the ISP-A path or ISP-B path?

Can you please tell me what is the mechanism used to select the flow path?

RBK

hclisschennai · ‎05-19-2009

Hi Joseph,

Thanks for your comments. But i find difficult to understand what you are trying to convey.

I will put it in simple ways. I have one Internet proxy server for outbound Internet connection with IP 218.248.6.2

I have 2 Internet ISP links A & B with A=4 Mbps and B=8 Mbps

When "max-range-utilization percent 15" under oer master what will happen with the traffic flows.

ip route 0.0.0.0 0.0.0.0 ISPB

ip route 0.0.0.0 0.0.0.0 ISPA

My understanding is Initiall traffic will flow through ISP B and if ISP B link utilization reaches 1.2 Mbps (15% of 8 Mbps), some traffic will flow through ISP A.

After this situation, how the traffic path is choosen. Whether forth coming flows are taken the ISP-A path or ISP-B path?

Can you please tell me what is the mechanism used to select the flow path?

RBK

Joseph W. Doherty · ‎05-19-2009

If you have:

ip route 0.0.0.0 0.0.0.0 ISPB

ip route 0.0.0.0 0.0.0.0 ISPA

and

max-range-utilization percent 15,

then (my understanding)

if ISPB had 40% load, ISPA could have a load +/- 15% or ISPA could have a load between 25..55% and be considered within balance.

Or, if ISPA had 0%, ISPB could have up to 15% (of B); conversely if ISPB had 0%, ISPA could have up to 15% (of A) and be considered balanced.

If links are not in balance (range), OER will try to "juggle" flows to attempt to bring links into balance. Often it does this by injecting more specific routes, at the aggregation you specify (/24 default?). Injected routes woull cause traffic that matches that aggregation prefix to chose just one path. Since OER can inject multiple routes, it balances the links.

For example, if you have 3 active flows going to a.a.a.a at 1 Mbps, b.b.b.b at 1 Mbps and c.c.c.c at 3 Mbps, it might inject:

ip route a.a.a.0 255.255.255.0 ISPA

ip route b.b.b.b 255.255.255.0 ISPA

ip route c.c.c.0 255.255.255.0 ISPB

which would load ISPA link 50%, ISPB link 37.5%; difference 12.5% - in balance.

PS:

OER can do more than just balance links, it can monitor actual performance (much depends on how you configure it). So, if in my example if OER notices traffic to b.b.b.b via ISPA performs poorly, it could move that traffic to ISPB. This would place the links out of balance, but provide better traffic performance.