What is big next year ?
Following from my predictions on what WILL NOT be big in 2012, finally, my predictions for 2012. Not early, but hey, the future never arrives on time anyway.
Although it’s called “cloud” most of the time, Data Centres have been mostly ignored for the last decade and most companies have truly obsolete infrastructure in place – all the way from power units, air conditioners, networks and servers, it’s all designed like Year 2000.
Using ten-year old switch infrastructure like the Cisco Catalyst 6500, obsolete load balancers 1 and manual device configuration isn’t going to carry the new systems requirements in the next decade. Even the introduction of blade servers was a hardware optimisation at best, and a vendor lock in at worst. All those custom backplanes don’t look so great when you want those new SSD storage caching cards in your servers
The virtualization push for operating systems is kicking Microsoft’s butt as the OS is no longer infrastructure, it’s an application that runs on the virtualization server. And the current networks no longer have enough performance to handle increased server density that this creates.
This is a universal truth for all data centres: last year your company has X servers, this year, virtualization means you can have X times two using more or less the same servers as last years. Networks and storage doubles but that not the server team’s problem.
Of course, the last decades under investment is this decades boom category. Virtualization drives just about everything now and is led by VMware, but that boom happened last year is now is in the mature phase. This year, VMware will do just fine, but will have less impact since it has competition from Microsoft, Citrix and some of the open source hypervisors. Importantly, EMC is tightening control over VMware and this will cause many customers to second source their virtualisation. EMC is not a supplier with a good reputation for sharing and caring about their customers, their shareholders matter much more (and all that means for customers).
To add a twist to this, you would think that building big data centres is new – it’s isn’t. The difference about some of the biggest data centres companies like Google and Yahoo can make cheap marketing and technical review out of talking about their data centres but secrecy and obscurity about data centres remains a vital part of the security for many companies.
Sometimes it seems like data centres were never invented until 2010.
Network Year after a Decade of Waiting
This year, the really big networking thing (out of many) is 10Gb Ethernet. Some companies, notably Arista, have claimed that the last two years are the year of 10GbE and probably are claiming it again this year — surprise. The is primarily driven by virtualization which is driving new server hardware that can handle more than a gigabit of bandwidth . High density servers, either as chassis like Cisco UCS or HP C-Class, or as bigger fixed format units like Dell’s PowerEdge 7xx/8xx/9xx means less physical servers which means less switches. The three tier or two tier data centre is finished and it’s now about flatter Ethernet networks with a new class of performance in low latency, greater throughput and more reliability.
The final phase will be the arrival of 10GBaseT “LAN on Motherboard” chipsets that will herald the dumb arrival of Cat6 copper cabling. Really, people, it’s time to give up on copper cabling – but the time you have and replaced your Category 5, upgraded to Category 6 and tested it’s easier and better to run a fibre optic cable, even a patch lead would be cheaper.
A secondary, but likely much larger market is Service Providers who will upgrade their backbone and distribution networks to use 10GbE to handle wireless and wired Internet growth. Also note, that Ethernet will continue to be the only Layer 2 protocol in the whole world. ATM/Frame Relay will continue to quietly rot in carrier comms room and they will ignore upgrading it for another year. (So what about lousy customer service and slow performance, just run MPLS over it and no one will notice. )
Merchant Silicon Drives Network Configuration Management in 2012
After years of reducing actual R&D spending, most of the established networking vendors aren’t truly capable of developing new chips anymore. It’s not an innovation, it’s a celebration of short term financial planning. You can accept the gloss from the vendors “that’s it’s not good business” to develop chips, but really it’s a failure of investment in R&D for short term profits.
The static nature of the networking business of the past ten years means that reusing existing chipsets (with upgrades or tweaks) with slightly better software or rebadging products from other companies has been the way forward instead of innovative and ground breaking silicon. While Intel shrank CPU fab size down to 45nm or less, networking chipsets continued at 200nm or 120nm processes until the innovation cycle forced new designs.
Even Cisco’s much marketed Nexus silicon is apparently running years late on expected delivery and the features that aren’t that far ahead compared to the merchant silicon.
It takes years to design and build silicon chips and decades to build teams that can do it reliably. I wonder if network vendors took short term profits in the last few years and many of those design teams transferred to contract companies in India, Taiwan, China and Indonesia which led to the decline and delays in new products. These same companies were subsequently picked up as contractors by merchant silicon makers who have done much better with the same resources, and producing higher quality chips than Cisco, HP or Juniper. That’s certainly what it looks like.
The market transition driven by virtualization has caught most networking vendors by surprise. For “innovation”" they have turned to off the shelf solutions from Broadcom, Fulcrum/Intel and other silicon foundries to power their current generation products with their existing software platforms and using their marketing capability to resell these products to their existing customers. The merchant silicon is the name of the chip families that provide true Ethernet capability and the old vendors simply glue these chips to a motherboard. The networking business now looks like Intel in the server marketplace and this will grow dramatically this year.
Is this a good thing or bad ?
In the sense that networking companies have failed to innovate and allowed other companies to make the market in switching silicon is poor business practice. A vertically integrated is inherently stronger (as Apple can demonstrate) but since networking vendors have only software on which to differentiate their products, customers are finally seeing user focussed features arrive on their switches and routers. In the past, the CLI and remote management features were rushed out the door and “fixed later”.
Of course, vendors will put a positive spin on using merchant silicon. Claims that “it’s let them focus on core business”, “lower production costs are good for customers”, “why build when it’s already available”are quite specious in my view. It’s just marketing spin on a bad situation.
Since software is the only thing that Cisco, Juniper and HP can use to differentiate their products, I’m looking forward to better features. To see that process in action, go and research Arista Networks. The only value they can add to their products is their EOS software platform which focusses on cool network operations capabilities. And networking stuff.
And it’s the features for network management that matter mainly so that we can have better software tools to configure and manage the network configuration. This feeds directly into the “Cloud” proposition since the whole purpose of the Cloud is to automate operational tasks. Many have tried over the last ten years, but the software APIs that are in SNMP are not rich or capable of anything more than performance and availability monitoring.
The EtherealMind View
As a result, I expect network vendors to make a big deal about their software platforms this year. Expect to see lots of announcements on XML APIs, support for cloud platforms like OpenStack, tight integration with NETCONF/YANG for configuration management, and the final recognition that SNMP isn’t part of the future. You’ll see lots of announcements about network management and operations lifecycle tools.
Also look for vendors attempting to build proprietary systems (likely called platforms by marketing knobs) in an attempt to lock customers in and/or lock competitors out. Any new technology tends to closed systems before customers force it into the open.
Make sure you pay attention to the fact that Merchant Silicon could drive a new wave of proprietary software solutions. We don’t need that in 2012.
- It’s surprising how many large companies still have Alteon (now Radware) load balancers from 2005 and earlier in the core of their networks. ↩