Advanced Computing in the Age of AI | Tuesday, March 19, 2024

Five Years On, UCS Makes Cisco A Systems Player 

Five years ago, when Cisco Systems launched its "California" Unified Computing System and entered the server business, that launch was not purposefully timed to come out into the gaping maw of the Great Recession. But it is probably a fortunate thing for the networking giant and systems wannabe that it worked out that way. Recessions have always accelerated transitions in IT architecture, and that usually plays in the favor of upstarts with new ideas and against incumbents who are set in their ways.

In this case, the benefits of the UCS converged system allowed the juggernaut in datacenter switching to not only get a foothold in the server space, but compelled IBM, Hewlett-Packard, Dell, and others to react by buying up their own switching businesses and forging their own converged systems to compete with Cisco. IBM may be in the process of selling its System x hardware business to Chinese PC maker Lenovo, which has systems aspirations of its own, but the company has no intention whatsoever to stop selling PureSystems converged systems against Cisco. IBM just wants to get out of hardware design and manufacturing and the low-end commodity X86 server business and focus on the converged value-adds that are representative in the UCS machines and, to be fair, were already on their way to be added to blade systems at the major server makers.

Looking back on it now, it is hard to remember how many people felt Cisco would have a tough time selling systems, even though with tens of billions of dollars in the bank it can pretty much afford to indulge in any technology it feels like. (Cisco has done this in the consumer space, with predictably bad results.) While the company certainly holds sway with its datacenter networking customers, the people who do server acquisitions at the large enterprises have their own preferences and prejudices when it comes to server iron. Server, storage, and network convergence for a certain class of workloads may seem normal now, but it was a new idea five years ago and one that could have been met with political resistance inside of datacenters even if convergence had technical and economic merit. But, as we pointed out above, recessions have a habit of making people focus on whatever new approach might save them some money over the short and long hauls.

"There was a lot of skepticism," recalls Todd Brannon, director of product marketing for Cisco's Unified Computing System line. "I think that people rightly questions whether we could cross the divide between the network buyers and have a conversation with the server buyers in the datacenter. It was definitely viewed as a pretty big step, and it changed a lot of competitive dynamics. And that was also when virtualization and cloud took off, and the needs of the IT team accelerated very quickly. People were fighting for survival, and more than ever, IT departments were under immense pressure to deliver efficiency and velocity for businesses to survive at the same time their budgets were getting slashed. It really was a valid question at that time to ask if this was the best time for Cisco to dive into a market segment with deeply."

The issue that drew Cisco in the server business was not just a desire to expand its share of the datacenter wallet, but that the company saw time and again that managing individual components in the datacenter was not a big deal – servers with or without hypervisors, storage internal or external, switching to link it all together – but that because of the nature of modern distributed applications, these needed to be integrated so they could be managed as a whole. This is why, as we have reported, that financial services giant JPMorgan Chase went with Cisco UCS machines for its massive 100,000-seat virtual desktop infrastructure rollout, and it is also why oil and gas explorer Apache Corp shifted to UCS iron for the clusters that run its reservoir simulations.

"Innovation had really stalled and had become a packaging exercise," explains Brannon. "We looked at how you would automate bare metal as well as virtual machines, which we do with our UCS service profiles. Then we have innovation on the virtual I/O side with our own ASICs and fabric cards, and we have been trying to dial up the conversations with customers that the way we abstract the bare metal is done in a way that hypervisors can't do it. We felt that innovation in the server business had stalled out, and with UCS, we focused on the operating expense issues that customers had."

Based on surveys of UCS customers, Cisco says that using UCS iron can cut server and network provisioning times by 87 percent, cuts out 77 percent of the cabling between devices, chops ongoing maintenance costs by 77 percent, and reduces power and cooling costs by 53 percent compared to rack-based pizza box servers and top of rack switches.

cisco-ucs-customers-over-timeCisco initially focusing on a very tight software stack consisting of VMware's ESXi hypervisor and vSphere and vCenter management tools combined with its own UCS Manager, which controlled the virtualized compute and network within the UCS system. And it kept a relatively simple product line, only adopting Intel's Xeons and forgoing chips from AMD. The tight focus, integrated and converged switching for both server traffic and server-storage traffic, and tight partnerships with EMC and then NetApp for storage made UCS a relatively safe bet. Before too long, pilots of one or two chassis became production systems with several racks of gear, and once one workload was put onto the machines, companies kept coming back for more to put other workloads on UCS machines. In many cases, large enterprises started out on UCS iron with their virtual desktop projects, then moved to test and development clouds and are now building private clouds and running big databases and applications on the heftier iron that is available today in both rack and blade form factors in the UCS line.

At this point, Cisco has over 32,000 UCS customers, including 75 percent of the Fortune 500, and importantly, more than 14,000 of these shops are repeat customers. The business has an annual run rate of more than $2 billion in Cisco's top line. The Virtual Computing Environment (VCE) partnership with EMC is driving more than $1 billion a year in Vblock system sales at the moment, which includes sales of UCS iron, EMC storage, various software, and services. The FlexPod preconfigured systems that Cisco makes in conjunction with NetApp are also bring in $1 billion a year in sales at the moment. These are the two big channel partners for UCS at the moment, of course, but Cisco has 3,850 channel partners pushing UCS machinery today and 80 percent of its sales go through those channels. IDC reckons that Cisco shipped 264,611 server nodes in 2013, a 26.6 percent increase over 2012 and making it the fifth largest maker of servers by shipments.

Incidentally, IDC also expects for these converged systems to represent 10 percent of server revenues by 2017, up from 4 percent in 2013. (That is the revenues for just the server portion of the converged system.)

Cisco generally has margins in the range of 65 percent for its switching and routing products, and while Brannon was not going to give precise numbers of UCS, he did confirm that the margins that UCS brings to Cisco are somewhere between those the company gets from switches and the relatively low margins commodity X86 server makers get. That extra margin comes from the fact that it is a converged platform with everything needed to more fully automate the provisioning and management of infrastructure for applications. Cisco supports Hyper-V and KVM hypervisors as well as the OpenStack cloud controller and the usual gang of Linux operating systems as well as Windows Server either as guests or on bare metal.

cisco-ucs-blade-share

Cisco made some waves when it acquired the WhipTail flash storage business last year, which it has woven into the UCS line and has said adamantly that it is not entering the storage business to compete against EMC, but rather adding flash to the UCS system to get it closer to CPUs and to accelerate compute. In the wake of VMware's announcement of its Virtual SAN (vSAN) last month, which as the name implies creates a virtual SAN across as many as 32 nodes that can be shared by the ESXi hypervisor and looks and smells and feels like an external SAN as far as applications are concerned, there are rumors going around that VMware and EMC are cooking up their own vSAN appliance, called Project Mystic. It is not clear if this appliance will be based on UCS machinery or on other system designs, but it stands to reason that it will probably be based on a minimalist server design, possibly coming directly from EMC and with an option for installation on UCS machines. Cisco could, of course, create its own vSAN appliance based on UCS even if EMC or the VCE partnership does not. Brannon was mum on the subject. But there is obvious appeal to get the storage, whether it is a vSAN array implemented on the servers or an Invicta flash array linked in from an external enclosure, tightly coupled with the virtualized compute and networking of the UCS machine.

The biggest change to the UCS setup, as far as Brannon is concerned, was the addition of the Invicta flash arrays to UCS earlier this year.

"We look at flash as a boundary technology," explains Brannon. "It has been used as a caching layer in storage, in servers, and now in DRAM. We have created a control plane between all of these elements, and our customers are asking us to please bring flash into the system and manage it in the pooled manner with service profiles. Our ultimate vision is to be able to use that flash to mimic storage or to mimic DRAM, depending on your application needs. This is not really a pure-storage play for us. Customers are going to need SANs and spinning disks for years to come."

EnterpriseAI