Packet inspection/traffic management

Caspian Soaks Up $55M More

Caspian Networks Inc. received a $55 million infusion of venture capital for its “aggregation routing” and traffic shaping business, bringing its total funding to date up to $300 million (see Caspian Gets an Extra $55M).

Caspian CEO Brad Wurtz says the new funding will be used to increase the company’s sales, marketing, and customer service presence in the U.S. and Europe.

Caspian’s business comes mainly from Asia (60 percent), with lesser portions from the Middle East and Europe (30 percent) and the U.S. (10 percent), a mix Wurtz would like to be more evenly distributed. “We are very well staffed in Asia, and we will be making the majority of our investment in the U.S. and in Europe,” he says.

Caspian’s VC investors bet its "flow-based" traffic management routers will be hugely appealing now that so many different types of traffic are traversing converged IP networks. Caspian's gear can, for example, allow a network operator to recognize and assign “best-effort” service to P2P traffic, while reserving “mission-critical” bandwidth for real-time applications like VOIP and interactive gaming.

Wurtz says he expects the company to reach profitability in the second half of 2006, “depending on the markets.”

Caspian began life as a core routing company, but later retreated from the space in the face of stiff competition from larger suppliers such as Cisco Systems Inc. (Nasdaq: CSCO) and Juniper Networks Inc. (Nasdaq: JNPR). The company now has focused its efforts on what it calls “aggregation routing” and its flow-state IP QOS systems.

“Our customers have told us that they want to use us in a different place in the network, at the aggregation points where they’re aggregating lots of edge devices and then shaping the traffic –- applying congestion control and QOS mechanisms –- before it goes over the MPLS core,” Wurtz explains.

But Caspian recently scored a substantial core routing win with the Korean government; and Heavy Reading analyst Scott Clavenna believes there may yet be room in the core router marketplace for Caspian (see Caspian Scores in Korea).

“The key thing to focus on is that core routing is actually coming back, globally, as operators move to VOIP and IPTV,” Clavenna says. “So they need real QOS in their core routers, and in some cases are looking outside the duopoly of Cisco and Juniper for that.”

But while traffic management is an important and growing focus in networks supporting next-gen applications, Caspian’s “flow-state” method of traffic management is just one way to do it (see Ellacoya Stands Alone).

“I do think flow-state routing is a very hot thing; the problem is there are some competing paradigms out there,” says RHK Inc. analyst Mark Seery.

Traffic management technology from Ellacoya Networks Inc. and Cisco use “deep packet inspection” to identify exactly what applications are generating traffic in the network. Caspian’s routers, on the other hand, take a more cursory look at the traffic, examining such things as its flow rate and packet size, Seery says.

“Ultimately I think the deep packet inspection that Cisco and Ellacoya are doing is what's really needed,” he says, adding that there is currently more market demand for deep packet inspection (see Ellacoya Sees Deep Packets at Shaw). “They [Caspian] are in a very interesting space, they’re in a hot space, but they’re really the only ones that are taking the approach they are taking.”

Caspian’s Wurtz says the new funding round comprised investments from the company’s existing VC partners, and was led by Oak Investment Partners, U.S. Venture Partners, and Morgenthaler.

"It’s unlikely they can recover the whole $300 million for their investors, but maybe they can get the $55 million back that they have just put in," RHK’s Seery says.

Caspian is headquartered in San Jose, Calif., and has offices in Research Triangle Park, N.C.; Tokyo, Hong Kong, Seoul, and London.

— Mark Sullivan, Reporter, Light Reading

Page 1 / 2   >   >>
turing 12/5/2012 | 3:13:12 AM
re: Caspian Soaks Up $55M More I claim that the provider has two choices: pay to provision about 10% more bandwidth, or pay for shphisticated QoS management. In general, I contedn that the bandwidth is a wiser and cheaper choice.
The "10%" is my made-up upper bound of the maximum theoretical gain you can get using the best theoretical QoS versus a trivial diffserv-based QoS in the core.

Maybe we're saying the same thing then. I had thought you meant just increase bandwidth vs. any QoS handling. But here you sound like you're saying it's not worth the extra to pay for more than just diffserv, right? (not sure) i agree with that.

My response to no Qos at all is that even when you look at a 15-minute peak window of say 70% link utilization, if you actually zoom in to the 50-100ms window intervals inside the 15-minute, there will be 100% line rate for the whole 100ms all over the place. But even more important is that is how the 100% bursts came about - by the aggregation of lower or same-speed links. Since the aggregation happens in somewhat random bursts, there are periods during which, for example, 2 uplinks will each be feeding >50% line rate and so their sum will fill the aggregate. Any more than 100% of the aggregate and there wil be congestion and if the router doesn't have big buffers there wil be loss. So the 100% you see on the core link is of delayed traffic and some lost you dont see. The only way to buy enough BW to get out of that happening is to buy enough for all the uplinks BW put together, which is far too expensive.

My only point here is every time i hear "its only 50% or 70% utilized" i cringe.
sgan201 12/5/2012 | 3:13:13 AM
re: Caspian Soaks Up $55M More Hi,

Firstly, I do not know whether Caspian router solve this particular problem that I am going to describe. But, so far all the discussion miss a very major problem in the core to access direction. Even if you make the core bandwidth as large as possible, you had just the traffic burst from core to access side worst. The core bandwidth is 10 to 100 times the access. The traffic burst will either overrun the buffer of the edge router (packet drops) or if the buffer of the edge router is very big, it will result in jitter and delay. So, there is value in have a box traffic shape the flow before it enter the edge. By doing this, you will result in less packet drop and better good throughput (goodput) of the network.

craymundo 12/5/2012 | 3:13:16 AM
re: Caspian Soaks Up $55M More As far as I understand the normal behavior of IP is to burst, so you need to add a lot of bandwidth in order to alleviate congestion.

"Bandwidth is cheap", it depends, on internet peering points and iternational links I dont think so.

Deep packet inpection boxes will achieve 10G? on a cost effective way will be difficult

Why people is more confident installing appliances on their netwoks and not better routers?
Mark Seery 12/5/2012 | 3:13:16 AM
re: Caspian Soaks Up $55M More Hi pro Caspian.

Caspian definitely has some interesting technology and as you rightly point out, there is every reason to believe it is more scalable today than deep packet inspection - that is a definite plus.

The question I think that is being missed in the dialog on this thread, is that the fundamental focus needs not to be on solving an end to end QoS problem (because no approach scales end to end today for microflows - and BTW, the comments that Caspian only operates in this mode is not true), but on unleashing revenue generating opportunities.

Some aspects of this can clearly be done with the Caspian solution, and perhaps in the short term, some of the most important aspects. The question you then have to ask is can deep packet inspection unleash a broader, more specific, and more targeted set of revenue generating opportunities. Perhaps it is the case that both approaches are complimentary. We will have to see.....

arch_1 12/5/2012 | 3:13:25 AM
re: Caspian Soaks Up $55M More No, turing, I'm not in fact assuming a TDM model. I am instead assuming that in the Internet core, the link rates are very fast with respect to individual demands, so the statistics are favorable. The utilizations I mention are peak utilizations, not average: assume I'm talking about the peak-15-minute utilization.

I claim that the provider has two choices: pay to provision about 10% more bandwidth, or pay for shphisticated QoS management. In general, I contedn that the bandwidth is a wiser and cheaper choice.

The "10%" is my made-up upper bound of the maximum theoretical gain you can get using the best theoretical QoS versus a trivial diffserv-based QoS in the core.

Note that the equatin changes dramatically on access lines. On access lines the cost of implementing fancy stuff is much lower, and the benefits are much higher, because of the much lower speed of the link.
3000psi 12/5/2012 | 3:13:27 AM
re: Caspian Soaks Up $55M More As a flow based system the system must scale to the number of flows that pass through it.

How many flows pass through an Internet core router?

How many flows doea Caspian claim to support at once?
turing 12/5/2012 | 3:13:28 AM
re: Caspian Soaks Up $55M More If a link is used at < X capacity, (say, X=.5) then QoS is irrelevant. Each packet will be forwarded with minimal delay.

No it won't. You're assuming a TDM model, which does not exist for IP, at the micro level (or even macro level). A very bursty nature with self-similar characteristics is what most experts believe actually occurs. I have not seen any recent studies though. (no one seems to do them anymore) But clearly some applications behave TDM-ish, like VoIP, but HTTP, bittorrent, email, ftp, and such all do not. There are bursts all the time, and they last for long and short durations.

Think of it as a highway, which is utlized at <50%, but long traffic jams occur all the time (like twice a day), as do small/short slow-downs.

Not that I think Caspian's model has any benefit for that over plain diffserv. I still don't really understand caspian's claim of differentiation and value over normal HW-based edge/core routers.
OldPOTS 12/5/2012 | 3:13:31 AM
re: Caspian Soaks Up $55M More arch_1 you make a good point. Why now would someone want great QoS. But consider;

My real life/measured experience is that the amount a very good QoS difference is about 20-30%. As you indicated the other side to this is how big is the cost of the core network capacity. Noting that the core to some people is a few miles of cable and to others it's a fiber across the country. A much different cost.

The relative cost of the more extensive core network for service delivery is about 30% access, 40% core, and 30% access. This only becomes important as the network link costs becomes the major contributer to the cost for delivering the services.

Right now for IP/Ethernet networks, things like all those servers and head-ends are a large portion of the network costs. And now the capacity generated by the subscribers is barely X <=.1. But later in the life of a network as those services are added, capacity requirements fill up the mature network's fiber capacity, and core costs become one of the few variables that can be reduced. Then someday in the future QoS becomes significant again. But time is now at internet speed.

That limit of fiber capacity is what drove the development for ATM (founding User member ATM forum), but fiber capacity improvements (like DWDM) lowered the price and the networks were operated @ X<=.5. However as the capacity filled up, a few years ago, I did many proposals that were comparing QoS more favorably than more network fiber/capacity addition.

The new networks will have to add those new killer apps that generate revenue and fill up the pipes to make good quality QoS costs valuable again. Especially since ports are cheap. Distributing traffic properly across a mesh network to save costs and provide availability, then also becomes important. (Current IP routing protocols don't cut it) I'll enjoy retirement on my sail boat till then and then roll out the answers for a price when it has an impact.

Also because of the burstiness of new application traffic and the sensitivity of video & voice to instant statistical peaks and delay, great QoS may become a needed marketing competitive requrement to maintain and acquire subscribers. So says many carrier LR surveys. Even a web page or attachment can contain a lot of bytes that the access can now deliver quickly in a GPON gig spike that interferes with those core converged continuous streams, unless these bursts are severly limited (scheduled) to statistically smooth their impact. I know it will happen someday, TM was my specialty.

But when will the networks fill up their capacity above X>=.6? Will I remain a retired sailor?

arch_1 12/5/2012 | 3:13:32 AM
re: Caspian Soaks Up $55M More If a link is used at < X capacity, (say, X=.5) then QoS is irrelevant. Each packet will be forwarded with minimal delay.

If demand for a link exceeds Y,(Y<=1) Then Best-effort customers are screwed.

This means that the best theoretical QoS can reduce the bandwidth requirement by X/Y. For the most conservative value of X and Y (X=.5, Y=1) the best theoretical QoS algorithm can reduce the required bandwidth by a factor of 2. for realistic values of X, and Y (X=.6, Y=.95) the factor is 1.35 to 1.

This describes the difference between no Qos and the best theoretical QoS. The difference between a trivial QoS algorthm and the best theoretical QoS algorithm is much worse: the best theoretical algorthim beats the most trivial ad-hoc algotorithm by at most 15 percent.

So, as a core Internet proivider, why should I pay more for better QoS? It's cheaper to pay for more bandwidth.

complex flow-based routing is only cost-effective if the bandwidth is extremely expensive. This may explain why Larry Roberts is now making presentations to the TIA satellite working group.
fiberous 12/5/2012 | 3:13:39 AM
re: Caspian Soaks Up $55M More Remember the VCs have access to retirement
monies and they still shake and produce
white ppt. Most of these blokes are very
Page 1 / 2   >   >>
Sign In