blogs.riverbed.com

Top 10 things Cisco must do to catch up to Riverbed

For the past five years, Cisco has been trying hard...real hard..to catch up to Riverbed.  It must be frustrating, because every time Cisco makes some progress in terms of introducing features, Riverbed seems to pull even further away with still more new capabilities.  To illustrate what I mean, I've compiled a partial list of things that Cisco still must do to catch up to Riverbed.

10.  Fix the WAAS per-peer data store and make it more like Riverbed's universal data store -- Unless they do this, WAAS simply won't scale as well as the Riverbed Steelhead solution.  Just imagine, in a network with 100 remote sites, the core Cisco WAAS device in the data center would store 100 times as much data as the Riverbed device in an equivalent deployment.  Here is a link to my previous blog on this issue:

http://blog.riverbed.com/2009/01/riverbeds-universal-data-store.html

9.  Get rid of the CIFS caching architecture -- The caching approach stores all CIFS data to disk twice, once in DRE, and a second time in the CIFS cache.  Can't be good for scaling performance when WAAS has to do roughly twice the I/O operations to disk as the equivalent Riverbed device in order to optimize CIFS performance.

8.  Grow WAAS revenues, not shrink them -- In the recent July (FYQ4/2009) quarter, Cisco's Application Networking business unit--the group that sells WAAS--shrank revenues by 27% year-over-year.  This comes on the heels of the prior quarter where the same business unit shrank revenues by 31% year-over-year.  Analysts such as Gartner have also noted a very significant shrinkage in market share by Cisco WAAS in the recent quarter.

7.  Optimize asymmetrically-routed traffic without relying on WCCP -- Let's face it, asymmetrically-routed traffic is common in large networks.  Cisco's requirement to use WCCP in such networks makes WAAS very difficult to manage and deploy, and is yet another reason why WAAS doesn't scale for large networks.

6.  Support something other than just Windows Core Services and ACNS on their virtual blade -- With Riverbed customers can deploy a full version of Windows, as well as Checkpoint, Infoblox, and a number of other 3rd-party vendor apps.  Many customers are also able to install their own custom applications into Riverbed's RSP and Windows running on RSP.  See my previous blog on RSP: http://blog.riverbed.com/2009/03/riverbed-services-platformworth-the-upgrade.html

5.  Improve software quality -- Even a feature-packed product is useless unless the software is stable.  According to Cisco's own product release notes documentation, the most recent WAAS 4.1.3 software has more "open caveats" (documented known bugs) than any previous major WAAS software release.  This can't be a good sign for a product that is supposedly mature or maturing.

4.  Catch up to Riverbed's 6500-customer installed base -- Cisco's most recent claim (Oct 2008) is that 3000 customers have purchased WAAS, and that number appears to include many who have since switched to Riverbed.

3.  Stop copying Riverbed and start innovating for once -- From RSP, to Steelhead Mobile, to NFS and SSL optimization, Cisco has always been in the mode of trying to catch up to Riverbed, by copying features and capabilities that Riverbed introduces first.  Admittedly, this is hard to do, especially when there continues to be a laundry list of missing features in WAAS that are available in the Riverbed solution.  But a sign of a true leader in any market is a vendor that innovates, and offers customers new capabilities that were not previously available.

2.  Demonstrate a truly large, enterprise-wide WAAS deployment that optimizes all TCP/IP traffic -- When you look into the details of each real-world WAAS deployment, they typically involve a relatively small number of WAAS devices deployed to support only a small portion of the customer's overall enterprise network, and/or a limited subset of the customer's TCP-based application traffic.  In contrast, Riverbed has many customers who have deployed hundreds of Steelhead appliances, including some customers who have Steelheads at every location in their enterprise network.

1.  Eat their own (WAAS) dogfood  If Cisco is selling WAAS to their customers, then they should be using WAAS in their own internal network.  And I'm not talking about just pilot testing.  While Cisco's internal IT organization is happy to use other Cisco products, they don't seem to be comfortable with WAAS.  My original blog about this issue (http://blog.riverbed.com/2009/06/rtp-eating-our-own-dogfood.html) has been met with silence from Cisco.  Since WAAS is not a new product anymore (Cisco has been selling it for almost five years), there is no excuse for Cisco not eating their own WAAS dogfood.

Since 2004 when Cisco acquired the original WAAS product, many have predicted that Cisco would catch up to Riverbed.  Year after year, Cisco promises that the next WAAS software release will truly be the "Riverbed killer."  And yet from all appearances it seems Cisco is further behind Riverbed than they ever have been in the past.

WAN Optimization and Thin Clients

"The era of the PC is almost over, and the era of the [thin client] is about to begin..."

Larry Ellison, March 8, 1996, in a speech to the Commonwealth Club of California

Well, it hasn't happened yet.  But even now, more than 13 years after Larry Ellison and Scott McNealy made their bold declarations about the future of network computing, there are still many who continue to see thin clients as the next major paradigm shift in computing.  Note that I'm not talking about just desktop virtualization as a tool--I'm refering to the utopia that they envision where all applications are centrally-hosted.  Perhaps it can still happen in the future, but we have learned quite a few things from more than a decade of attempts at applying thin-client principles to real-life computing requirements.  More likely than not, the future will involve at least some involvement from full thick-client PC's, if not continued dominance of the prevailing client-server computing model.

In a previous blog, I described how Riverbed is able to optimize the transfer of network traffic from thin client computing platforms.  But on the other hand, the advancement of WAN optimization technologies over the past few years now affords new break-through solutions for remote computing, where the thin client approach was previously believed to be the only viable option.  Could it be that maturation of WAN optimization technologies will not only preserve the trusty client-server computing model, but also threaten the bottom line of certain vendors such as Citrix who continue to yearn for that thin client utopia?

Both WAN optimization and thin client computing have common goals of centralizing and consolidating servers into the main data center, but that is about where their comonality ends.  Though WAN optimization is a more recently-available technology, thin client computing approaches have been explored and investigated for more than a decade.  And one of its most glaring disappointments has been the high cost.  Originally, thin client technology was supposed to be cheap.  That was one of the original attractions--deploy low-cost hardware at the branch offices.  But as we all know now, those Citrix software licensing fees are very expensive and a huge cost burden to many IT departments.

After 5 years at Riverbed, I have become somewhat jaded by the impressive reports of fast LAN-like performance for CIFS, Exchange, NFS, HTTP/SSL, etc.  All Riverbed customers experience these benefits, and I am not the least bit surprised by these reports.  But on a recent trip to Asia, I ran into multiple customers who are now telling me that they are also removing Citrix XenApp licenses from their branch offices that receive Steelhead appliances.  In fact, one Riverbed customer is removing Citrix XenApp for one critical application used by hundreds of users at 80 remote sites.  They expect to save $millions over the lifetime of that application, from not having to pay Citrix for annual recurring licensing costs that seem to rise faster than the rate of inflation.

While in the past I've heard of a few Riverbed customers removing Citrix licenses before, I am now starting to see it happen with greater recurring frequency, on a much grander scale.  It seems that WAN optimization has now matured enough as a technology that it can now be trusted for deep-impacting IT initiatives.  At Riverbed, we especially look forward to this new and interesting competitive frontier.

Are all WOCs made alike?

WAN optimization vendors seem to be saying the same things.  They all seem to be making claims of 50X performance improvement.  They all seem to talk about how users can benefit from LAN-like performance.  They all seem to claim that their products facilitate server consolidation and data center elimination.  As a concrete example, below are two graphs taken from marketing materials published by two competing WAN optimization vendors.  The first graph was originally published by Riverbed in 2005.  The second graph is from Blue Coat, and it is dated from 2007.  As you can see, there are some striking similarities in the claims by each vendor.

 

 

The central goal of vendors such as Blue Coat who mimic Riverbed's messaging is to make their products appear to be same as the Riverbed solution.  And by using carefully-controlled testing procedures in isolated lab environments, they often can generate performance results that appear to be Riverbed-like. 

But the fact is that competitive products are very different from the Riverbed Steelhead solution; they were created with very different product architectures, and often were originally designed for entirely different purposes (e.g., web security or file caching).  They also use legacy caching-based or tunneling-based architectures that have been tried many times in the past--with limited success.   In contrast, the Riverbed Steelhead solution uses a unique TCP-proxy approach that is innovative and purpose-built specifically for WAN optimization. 

Differences between Riverbed and competitive products become evident in real-life network environments.  Issues that are never observed in the lab expose themselves when the WOC is subjected to heavy traffic loads and network events that occur in a real-life network.  Typically, these problems are related to scalability issues.  For example in the case of Cisco WAAS and Blue Coat's ProxySG, both products suffer from scaling limitations related to their use of per-peer data stores (see blog Riverbed's Universal Data Store).  This is but one of a number of different scalability issues that commonly surface when the products leave the sheltered environment of the test lab.

Riverbed's competitors would like nothing better than for their prospective customers to limit product evaluations to only simple lab tests, where their scalability issues are not exposed.  To avoid falling into this trap, I would advise the following as part of the evaluation process for any WAN optimization product, to ensure that the scalability of the product is properly investigated:

• Do not limit your evaluation to just simple lab tests.  Always do pilot testing with the actual products in your production network.  Don't accept excuses from the vendor for not providing you with equipment for a limited time to perform the POC testing.
• For large networks, the POC should involve multiple sites, certainly more than two. As much as possible, test an environment where multiple branch offices access data through one core data center device.
• Test a wide variety of different applications that are found in your network, and under heavy traffic loads for both optimized and pass-through traffic.
• Monitor for negative impacts to non-optimized pass-through traffic.  Be aware that many WOCs gain performance benefits by stealing bandwidth from non-optimized traffic.
• Find out if the vendor eats their own dogfood.  Does the vendor use their own WOC product at all locations throughout their internal network environment?  If the vendor can't scale their own product in their own network, then how can you expect to in your network?
• Most importantly, ask the vendor to provide contact information of at least two customers who have deployed the number of WOC devices that you envision for your network, and speak confidentially with those customers about their experiences.  Has the customer realized real $$$ benefit from server and data center consolidation as a result of using the WOC product?

Despite what some vendors would have you believe, the bottom line is that all WOCs are not made alike.  I have met and spoken with a number customers who tragically found this out too late.  Some were distracted by lower prices offered by their originally-selected vendor; others by vendor/brand loyalty.  Each case resulted in many months or years of frustration before Riverbed finally came into the picture.

It’s a Mad, Mad, Mad, Mad World… so replicate, replicate, replicate, replicate.

Ok, I may be dating myself here but have you seen the 1963 United Artist's movie “It’s a Mad, Mad, Mad, Mad World”?  One of my favorite scenes is this:

Emmeline Finch: “Oh Russell, I feel sick.”
J. Russell Finch: “Now take it easy honey, these things happen ya know.”
Mrs. Marcus: “Now what kind of an attitude is that, these things happen? They only happen because this whole country is just full of people, who when these things happen, they just say these things happen, and that's why they happen! We gotta have control of what happens to us.”

Apparently you can divide people into two types, those that say “these things just happen” and those that “gotta have control.”  Now suppose you’re responsible for disaster recovery or business continuity, which type do you want to be? 

If you’re in the “it happens” camp, could you please send me a note of your company’s ticker symbol so I can short your stock?  I’ve been looking for a good tip.

If you’re in the “have control” group, you probably get that the basic principle of data protection is take everything you need and copy it somewhere else, far enough away that it’s there when you need it.  You might also appreciate that your wide area network is a handy piece of infrastructure for moving all this data.  In fact, I’ll bet you’ve already come up with a means of protecting your business critical systems, and you might well be performing some pretty high end data center to data center replication.  Like using HDS TrueCopy, Universal Replicator, or NAS replication, for instance.  Good for you.

Now, maybe that’s still not enough.  The WAN might not have enough bandwidth, or might have too much latency, or might have other applications running on it.  Lucky you, Riverbed WAN optimization can get you out of this bind, vastly improving the performance of long distance replication.  And even better than that, Hitachi has now qualified Riverbed for use with their replication tools.  This joins our EMC SRDF/A qualification and ton of other storage partnerships, and all of these arm you with many great ways to get the disaster recovery plan you want.

Hope is not a strategy.  Be prepared.