Advanced Computing in the Age of AI | Thursday, March 28, 2024

VCE Vblocks Chase More – And Bigger – Workloads 

Partnerships in the IT sector don't always work out, but it looks like the Virtual Computing Environment effort between server upstart Cisco Systems, storage juggernaut EMC, and the latter's VMware server virtualization subsidiary is. Large enterprises are moving from proofs of concept to mission-critical workloads with Vblock systems because they want a new – and less costly – way to build and manage physical and virtual infrastructure.

The company has just rolled out a new member of its Vblock converged system family, with Cisco Unified Computing System server nodes sporting Intel's Xeon E5-4600 processors plugged into EMC's shiny new VNX storage arrays and VMware's new ESXi 5.5 hypervisor. The new setup, called the Vblock System 340, is a two-rack system that sits between the existing one-racker System 200 and the three-racker System 720, and again is unique in that it has the latest hardware and software available from the three partners. VCE is also announcing two specialized appliances that are tuned up specifically for Oracle and SQL Server databases or virtual desktop infrastructure (VDI) workloads. (VCE launched an appliance for running SAP's HANA in-memory database back in February.)

VDI has been a key workload in getting VCE's Vblocks a beachhead in the data canter, Todd Pavone, executive vice president of product development and strategy at VCE, tells <i>EnterpriseTech</i>. And the latter is becoming more important as companies who tried out Vblocks for greenfield jobs like VDI look around and start adding more systems, ripping other workloads off of existing X86 systems and into Vblocks.

The VCE partnership hit a $1 billion annual run rate for its revenues back in February, and in the second quarter, it was growing at 50 percent, matching the growth of Cisco's overall UCS server business. (Cisco's UCS business has a $2 billion annual run rate, and that does not include external storage as the VCE number does.) Pavone says that VCE has deployed over 1,200 Vblocks to date – that is unique systems, not a rack count, and a system can span many racks and often does – with many hundreds of thousands of server nodes humming away at its 600-plus customers. These tend to be larger companies, of course, and Pavone brags that 8 of the top 10 telecom companies in the world have Vblocks as do 3 of the top 5 big pharma companies. Global expansion is being done through partnerships with Softbank in Japan and Capita in Europe.

"The growth and the adoption across all markets, in particular in Europe, we are seeing incredible growth," says Pavone. This in a server market that is flat or declining.

What is driving such growth? The desire to integrate and automate.

"The problem that we are solving for these big customers is that the vast majority of their operating budgets – well in excess of 70 percent – is spent on maintaining operations," says Pavone, and this is a number that we have all heard for the past decade and, until now at least, no one has been able to move the needle much on it to bring it down. "CIOs are getting beat up on budgets, which are flat or decreasing, and they look at their data center and they see silos of technology and silos of people and they are looking for a transformation. Ultimately, they can't keep buying technology the way they have always bought it. They can't buy reference architectures because it is not going to transform their operations. It may simplify a cap-ex buy, but it is not going to simplify operations. They are looking for a paradigm shift."

And, as it turns out, taking that operational portion of the IT budget down to somewhere between 30 and 40 percent of the total pie, which is a target that VCE and its parent companies say they can meet. The savings can be reinvested in IT or put to other uses in the company.

VCE started shipping Vblock systems at the end of 2010, and got its first phone calls when large enterprises were doing a greenfield project and were willing to give something new an honest try. In a lot of cases, this was a VDI proof of concept or an upgrade of Exchange Server. Then, they come back again and start buying Vblocks as part of an upgrade to their core Oracle or SAP applications, and then they return to bring other legacy workloads over. Depending on the quarter, anywhere from 50 to 70 percent of the revenues that VCE is generating is repeat business from existing customers, says Pavone. "For us, that means we are doing something right. They are seeing the value of it, and it is quantifiable."

VCE had the analysts at IDC poll its customer base and try to put some figures on the benefits they are getting from converged infrastructure. Across five dozen customers polled, customers, on average, were able to deploy raw infrastructure four times faster (40 days instead of 160) and it took 79 percent fewer people to configure, test, and deploy it. The annualized costs of that infrastructure were half that of the more traditional (and I presume unvirtualized and unconsolidated) rack-based server infrastructure it replaces. To be more specific, those using Vblocks cut their storage costs by 50 percent, network hardware costs by 40 percent, server count by 66 percent, power consumption by 16 percent, and data center space requirements by 12 percent. IT staff costs fell by 38 percent on Vblocks compared to prior infrastructure, which is sure to catch the attention of more than a few CFOs.

In the past year, the number one workload on new Vblocks sold was mixed X86 server workloads consolidated onto virtual machines, which reflects the growing diversity of jobs customers are putting on the systems. If you tease out a single new workloads put onto Vblocks, then the most popular one is VDI, and the next largest driver of business is the Microsoft Windows server stack, mostly Exchange Server, SQL Server, and SharePoint. Oracle databases and applications are next in popularity followed by SAP applications. After that, it is mostly custom, homegrown apps.

Of course, by going with a converged system, you give up some freedom. You can buy FlexPod converged systems, which mix and match Cisco UCS iron with NetApp storage, or you can buy raw Cisco servers and integrated them with other storage if you want to. Enterprise shops are used to best-of-breed. The important thing for VCE is that many shops already have chosen EMC as their storage provider and VMware as their virtualization and cloud orchestration layer, and Cisco for their networking. So the only real change is Cisco on the servers, and one out of four elements is not the same risk as two, three, or four out of four. This is one of the real reasons why Vblocks have a pretty decent run rate.

The Vblock System 340 is based on the blade, not rack, servers in the UCS family, and in particular has the B420 M3 blade. This is a four-socket machine based on Intel's Xeon E5-4600 v1 processors in the "Sandy Bridge-EP" family, which have not yet been updated to the Ivy Bridge designs with a 22 nanometer process shrink. (And it is not yet clear when that will happen.) You have two UCS enclosures (with a total of eight blades each) as a minimum configuration and the maximum number you can add depends on which EMC VNX storage arrays you choose. The fattest Vblock System 340 has 128 blades with EMC's VNX5800, VNX7600, or VNX8000 arrays, and EMC's Fully Automated Storage Tiering (FAST) cache, which uses memory and flash in the storage array controller to speed up data accesses, is enabled on the Vblocks. The Nexus 1000V virtual switch from Cisco, which plugs into the hypervisor, is also part of the stack and so is the MDS 9148 Fibre Channel switch from Cisco to talk out to storage arrays. You can use the Nexus 3048 switch with the machines as well as the integrated switching in the UCS chassis to hook the nodes to each other in a cluster.

VMware's vSphere Enterprise Plus 5.1 variant of the ESXi hypervisor (the one from last year) and the 5.5 release (which was just announced in August) are both supported on the server nodes. The management stack for the boxes includes UCS Manager for the server and networking, vCenter Server for the hypervisor and virtual machines, and Unisphere Manager for the storage. The setup also has an updated VCE Vision monitoring and logging tool and can do push-button updates for patches of systems software on the boxes. It includes an OEM version of vCenter Operations Manager, too.

The upshot of all this is that the Vblock System 340 has about four times the compute performance and twice the storage capacity as the System 300 it replaces. Pricing was not set at press time, so it is not possible to do the price/performance analysis at the system level.

(One of our projects here at <i>EnterpriseTech</i> will be to do a more thorough price/performance analysis of the Vblock line.)

In the fourth quarter, VCE will crank out two more appliance versions of the Vblocks, what it calls Specialized Systems.

The first will be a configuration that can handle millions of I/O operations per second and will be tuned for OLAP and OLTP jobs running on Oracle databases. This will eventually support IBM's DB2 and Microsoft's SQL Server databases, too. On the VDI front, VCE will be forging a Specialized System for Extreme Applications, which will have all-flash storage and inline de-duplication to make virtual desktops perform better and yet take up less storage back in the data center.

The feeds and speeds of these two Specialized Systems configurations were not yet divulged. We'll track them down as soon as they are out.

EnterpriseAI