<<   <   Page 2 / 2
ozip 12/5/2012 | 2:58:09 PM
re: DPI Market Set for Mobile Ramp Not sure anyone noticed but most of the boxes used to terminate mobile data already do the generalized equivalent of DPI (eg Starent).

DPI is to general a term however that changes little regarding this market, is it better to invest is speed or control. At this stage, consumers have been well trained to evaluate operator service offering based upon the speed offered. Its very unlikely that will change. As we saw recently with a large cable operator, claiming that they were messing around with packets to improve customer experience didnt wash.

Turns out that the individual boxes that we call DPI are deployed more to do man-in-the-middle modifications to specific protocols. As we also recently saw, these types of bandwidth management techniques can go badly wrong. They are both complex and likely to fail as the protocols they are dissecting change.

Before the DPI business gets much bigger, the valuable features of DPI will have been identified and put back into the router/switches that directly manage the data forwarding path. The only hope for DPI is in security, but there are lots of firewalls already (remember firewalls, functionally pretty similar)

OldPOTS 12/5/2012 | 2:58:04 PM
re: DPI Market Set for Mobile Ramp ozip,

Please see delphi's #9 post on QoS and DPI first.

At the edge, were most of those cheaper Ethernet switches are used, one must do DPI (more than a couple of bits) to categorize the traffic to do QoS within a congested networks. As mentioned categorization with DPI has trouble keeping up with protocol changes. Subs pay cost for general DPI QoS service in congested network.

To efficiently solve restricted resources (network congestion) one must pre-allocate shared resources and costs to various users at congesting nodes. This has two drawbacks; dynamically managing the resource allocation across the entire shared network and the complexity/cost at the nodes, especially at the edge router/switch. I believe that both of those have mostly been overcome with dynamic allocation protocols (thru UNIs) and modern faster and cheaper node processors.

Network operators must make a choice of where and how they spend the money in the network (links and nodes); see 'PBT Cost Claims Questioned'. And initial ARPU is important to getting the business case started. But success means either faster links or smarter nodes closer to the subscriber to avoid subscriber congestion.

Several years ago several cisco employees recognized that IP costing models had problems of distributing resources (All links and All nodes in a network) so they and many other users and vendors worked for several years to create a standardized protocol that supported grandular QoS. Before IP CoS (less granular) for those with a long memory. Sounds like something similar will be coming back soon because of successful congestion. Has one solution really disappeared to the back room awaiting revival, now that the problem is becoming apparent?


(Hint: 'Fast Packets' preceded the protocol)
VBNs forever!
lostinlight 12/5/2012 | 2:58:03 PM
re: DPI Market Set for Mobile Ramp In the late 90s, there was a lot of activity on QoS and efficient traffic shaping. After a while, people just threw more bandwidth at the problem. Also, the internet didn't really grow at 10,000% per minute as they predicted. So it has been working for a while.
Remember stardust.com?

With limited bandwidth in wirless, we may have to resort to traffic policing.

<<   <   Page 2 / 2
Sign In