Redundant Array of Inexpensive ISP's?

[Please reply off-list. I'll summarize back to the list if there
is more than a little interest in me doing so.]

I'm curious if anyone has experience with products from Talari
Networks, or anything similar, and would like to share. Did they
live up to your expectations? Caveats?

This seems similiar to Cisco performance routing.

See for more.

Tim Utschig wrote:

Good question. I'm also curious if anyone has experience with the
Mushroom BBNA device and how it compares to Talari. Due to various
premises and telco issues, I've been unable to get anything faster
than a DSL connection pulled into a certain branch office. I'm
considering any and every alternative at this point.

The Talari device appears to operate like the old Routescience
Pathcontrol BGP load balancer circa 2002 (Routescience is now owned by
Avaya I believe). Routescience was able to compile the best path to
Internet BGP prefixes so that a web site could connect to multiple 2nd
tier ISPs (for circuit cost and redundancy reasons), and control the
Mbps traffic over the best path, irrespective of the BGP feed supplied
by the upstream ISP. In my experience devices such as Routescience
automated the tedious work of using CAIDA tools to manually calculate
the best BGP path to destination prefixes, and eliminated the almost
daily reconfiguration of BGP route maps on Internet border routers.

Routescience was a great product that put dashboard BGP routing control
in the hands of the network engineer, and saved MRC circuit costs to pay
for itself within a few months.

Yes and no.

Yes, in that it does best path selection, no in that it does not use BGP, since low cost assumes DSL or cable, over which I've never seen BGP deployed. This class of device assumes an appliance at each end. Performance data is collected, compression and load balancing techniques applied, and a sum total improvement of capacity and reliability is achieved.

If you notice similarity between Talari and Route Science, they both do cover the same field and by several of the same people. Also, there are several producers of BGP route selectors, netVMG (now FCP at InterNAP), Proficient Networks (InfiniRoute) and Cisco's follow-on product.

I have no direct experience with Talari appliances.


Hello Tim,

a lot of our customers need a very stable Internet access got their
portable address space and their AS number from us (we are a LIR) and
connected to 2 or even more upstreams.

Sure, some of broadband ISPs didn't provide BGP for their clients, but
there are companies providing BGP over L2TP or GRE.

So all the solution costs ~$1000 one-time fee (PI/AS, BGP router like
Cisco or Quagga box, a bit consulting).

Good advice is to diverse upstreams by the media, i.e.
CaTV+DSL+Fiber+Radio, so if fiber to the house is cut - radio still working.

It is possible to integrate that to a complete service - i.e. install a
box that connects to 2-3 ISPs and "just works", but we haven't requests
to to that. Please, contact me off-list if somebody interesting in it.

Tim Utschig wrote:

Tim Utschig wrote:

[Please reply off-list. I'll summarize back to the list if there
is more than a little interest in me doing so.]

Please do. There are many rural ISPs and WISPs that might benefit from a decent look at these products, or any open source clones that might be available to test & refine these tricks.

Pricing for even a fractional DS3 in the rural US is still very high. Being able to shift bandwidth from a colo facility in a large city to a remote site served by 3 or 4 consumer grade broadband links could be a helpful development, if the bottom line works out.


In answer to a question below about experience with similar products...
Cisco IOS has the dynamic routing injection feature as part of recent
IOS versions.

The feature is now called Performance Routing (PfR) formerly known as
OER (Optimized Edge Routing) and as of 12.4(24)T, it can optimize
routing protocols other than BGP or static routes (called PIRO Protocol
Independent Route Optimization), including IS-IS, OSPF and EIGRP. RIP
folks should learn about routing protocols :smiley:

PfR does not do compressions/tokenization of the data, so it has no
Caching/compression/WAN Acceleration features, BUT it does do dynamic
path re-routing based on your policy or observed metrics like latency,
packet loss, jitter etc and can also do it based on observed Netflow
data and automatic instatiation of IP SLA active probes to see what
happens for a RTP data stream marked with dscp 46 or video stream
marked with dscp 34 and so on. As of recent IOS versions (12,4(9)T + I
think), it can control both inbound and outbound directions, and can do
things like send your traffic to ISP X up to bandwidth Bx and then shift
traffic over to ISP Y up to bandwidth By to do dynamic load sharing of
traffic to IP transit commit levels.... Not a bad feature for free.
Larger scale deployments should probably use a dedicated controller box
making the re-routing decisions, but any WAN egress point to an Internet
or private WAN provider is your "border" device used by the "master" to
get information, setup probes and learn netflow data to make decisions.

I've used it for testing purposes on enterprise WAN deployment and it
works pretty well. We are planning on deploying on a production DMVPN
solution when the MGRE bug below is resolved. My main beef is a bug
related to use of PfR on mGRE tunnel interfaces and the memory-hog
nature of the feature... It will detect your brown-out issues like
increased packet loss for traffic through provider X that cause
customers to call you about broken applications and will re-route the
traffic so you may never even know there was an issue!! The solution is
particularly good for enterprises with only a few WAN or Internet exits
from a location and for dynamically load sharing traffic to paid-for
commit levels to reduce recurring cost and get the most out of existing
connectivity without paying burst charges. We've done testing on use
for our internet border routing in the "advice" mode, where is just says
what changes it would maek, without actually making the changes.
Production deployment soon as part of the ever popular cost-reduction
efforts currently in vogue in enterprises right now given the current

There's some similar solutions out there.. RouteScience was mentioned,
but I didn't see anyone mention InterNAP FCP, which is part of the basis
for InterNAP's PNAP business model... They also sell it to others
enterprises and ISPs.