New OS, New technology
It is a given that this product has been extensively tested, and will work as described, probably with a small number of the usual glitches. Cisco commitment to software testing that commenced around 2004 / 2005 has definitely reduced the number of significant problems in the field. I do not believe that we will see repeat of the IOS 12.1.5T where so many features were introduced and took years to finally stabilise. Notwithstanding, NXOS is new and needs careful consideration. Currently, I am most concerned about integrating NXOS into my current NMM platforms. I suspect that Cisco will have addressed this at some level by deliberately maintaining SNMP MIBs and the IOS CLI but NXOS will have different considerations for CPU, memory, and other key monitoring variables that will take some time to adopt.
Fibre Channel Over Ethernet (FCoE)
The move to integrate FCoE into the Supervisor is not entirely unexpected. It is in Cisco’s interest to leverage their strength in Ethernet switching to extend the market share of the SAN market. However, what is striking is that Cisco has delivered a FCoE solution before the standard is completed. But Cisco has made this a key part of their marketing message.
Cisco is a regular player in the standards bodies, and we have seen Cisco deliver technologies such as HSRP and ISL, long before IETF / ANSI / etc have even started standards process. The key difference here is that no interoperability was ever needed. Thus HSRP only worked on Cisco Catalyst switches.
FCoE requires to co-operation of companies who are notoriously closed organisations. Indeed, Fibre Channel SAN can be viewed as a move by Storage vendors to make sure that they owned the entire solution to maximise their control of the customer (probably to the customer’s cost). Is it reasonable to expect that EMC and NetApps are willing to forgo the revenue of Fibre channel adapters and switches ?
Approvals and Validation
One of the greatest mysteries to me, is how Storage Vendors have convinced their customers that validation is a prerequisite for support. This seems an archaic concept harking back to the halcyon days of mainframe, and definitely not for the better.
Every storage equipment purchase is regularly vetted by the vendor (or worse, vendors) and long lists of ìapproved equipmentî are consulted prior to any purchase. And yet, in networking, we have no requirement for this process. Our standard are clearly defined, and the market quickly disabuses vendors from introducing incompatible features.
Quite why testing and validation takes months is also a mystery to me. The acquisition of VMware by EMC has seen a marked rise in this problem. Not too long ago, VMware was used on any equipment, and supported by VMware.
Is Cisco going to gain these approvals in time for us to use Fibrechannel ? Are EMC / NetApps / VMware going to support them ? Also, will there be quality drivers for Windows Server 2008 and 2003 ? These unknowns are a concern to me.
Networking versus Storage
For very large data centres, the Nexus will take a long time to reach acceptance. The merger of Storage and Networking people into a single team will take some time due to human resistance to change, and the the changes to operational procedures will take a long time to achieve.
Storage people will have a lot of personal investment in storage products both in terms of experience and certification, and professional pride. Also, they are resistant to change, some pockets are deeply hostile to new advances.
The critical nature of storage means slow take up of the new. Comments such “we trust EMC” or “NetApps is proven reliable” and “compliant” and “certified” are often heard.
While I agree that this has happened before with TDM Telephony to VoIP, this is no guarantee that the convergence of storage and network is a foregone conclusion.
Nothing to Lose
It is worth noting, that due to the architecture of the NXOS a a fully virtualised core, that if FCoE doesn’t gain acceptance, then Cisco can just remove it from the software. The software will still be a monster Ethernet switch platform and Cisco will only lose the investment in FCoE not the entire platform.
Penetrating the Server Farms
Still this represents a major shift for Cisco, which has struggled to make its presence felt in the Storage space (although they claim to have 50% of the space that VSAN space that MDS occupies). Since SAN are largely the preserve of the server teams, Cisco – as a network company – may have difficulty gaining mindshare. I note that Intel has put a lot of weight into FCoE, so Cisco is not alone here.
Early Upgrades – how many ?
The first versions of the Nexus 7000 have a capacity of 230Gbps per slot, but the total slot performance is an order of magnitude larger than that. Logically this means that Supervisor and Line Card upgrades are going to be part of the lifecycle to reach to higher performance.
This isn’t new. People who purchased the first Catalyst 6000 units have had eight or nine upgrade cycles to reach the maximum capacity of the chassis. In 1999, the first Supervisor 1 supported switching only and had a 32 gigabit per second shared backplane. It took a few years before we had the fabric switching modules and routing supervisors (using feature cards) and five or six years until we got to the 20 Gbps per slot capacity that we have today, and the first high speed modules were really expensive.
The most important fact that I remember, was the first two or three years was an endless cycle of upgrades to get the basic features that I wanted. As an old (not bold) engineer, I am wary of entering this cycle again.
Missing Features – so many….
There are so many features missing in the product. MPLS is one the biggest, you need to look really carefully to realise that the current version does almost nothing but switching and the most basic routing. Once you realise this, you can only position the NX7000 as either a high speed ethernet switch in your data center core for terminating 10GE, or as an access switch for servers that are using 10GE where, hopefully, you might even be using FCoE. Spending several hundred thousand dollars on an access switch is going to be difficult.
Splash the Marketplace
If you accept my previous comments, then I believe has Cisco released the NX7000 following reasons:
- to head off the entry of other 10GE vendors such as Juniper, Woven and Foundry to their customers. ( This also will scare off further venture capital funding on 10GE )
- to put the product into the design pipeline of corporates and service providers, in the expectation that deployment will probably start in two to three years, by which time the features have been delivered.
- to show commitment to FCoE and provide marketing impetus. Some customers will assume that if Cisco does it, its going to be big (and most likely they will be right).
- to shake up the storage marketplace, and get the customers asking the Storage companies to release FCoE products, with approvals!
- because the C6500 is showing grey hairs. The recent VSS upgrades don’t improve performance, just extend the life of the current solution. More upgrades!
- Fiber channel adapters and switches consume extravagant amounts of power( a big weak point). By laying out the FCoE commitment, funding for new hardware programs on FC for power reduction will be less likely.
Conclusions
It is probably too early for most of us to rush out and procure the Nexus 7000, it costs too much, it needs time for management tools to mature, and it isn’t a ‘must have’ right now. I suspect Cisco is marking out territory to prevent competition. I am unconvinced the FCoE will change the world, and concerned that many features are missing.
I think the product needs quite some time to mature before I will be ready to take the plunge.
