How EMC's World Pulls the Data Center Together

by Drew Robb

Hard-Core Hardware: To EMC, VMware and Intel chips are the glue joining storage and server hardware.

Last week's EMC World in Orlando painted a picture of the near future — one that includes a steady convergence of storage and server hardware, as well as the integration of virtualization into the fabric of the data center. EMC sees all of this convergence as hinging on VMware and Intel processors.

It is the advent of the Intel Nehalem chip, in particular, that has caused EMC to drop its long-term reliance on ASIC chips for its higher end Symmetrix disk array line.

"Nehalem is now a major processor for both servers and storage," said Greg Schulz, an analyst with StorageIO Group (Stillwater, Minn.).

Joe Tucci, chairman, president and CEO of EMC, (Hopkinton, Mass.), is firmly in the Intel camp. He said there are few if any workloads Nehalem can't deal with in a heavily virtualized environment. After decades on other platforms, especially for its flagship Symmetrix products, it has thrown in its lot with the chip giant.

"We are focused on the x86 architectures, as it accounts for 95 percent of shipped servers and 65 percent of server revenues," said Tucci. "it is also growing faster than any other chipset. However, I don't intend to get into the core server business."

EMC's latest and greatest, the Symmetrix Virtual Matrix (V-Max) is on x86. Compared to the Symmetrix DMX-4, it provides three times the performance, many more IOPS per dollar, and 95 percent faster provisioning. It also consumes 20 percent less power. While the DMX-4 has 8 ASIC chips, the V-Max takes that down to one Nehalem.

The company's grand plan is to connect the V-Max engine to x86 servers to provide the storage backbone for thousands of virtual machines (VMs). As the V-Max can scale up to eight engines and scale out to 256 engines, this equates to tens of thousands of disk drives, hundreds of thousands of VMs, tens of millions of IOPS and hundreds of PBs — all federated into one virtual matrix. Thus, all the servers can see all of the storage within one large virtual environment. This could potentially achieve the elusive goal of storage auto-provisioning across a massive environment. According to EMC, 96 percent fewer clicks are needed on V-Max provisioning compared to a traditional SAN.

VMware Glue

The star of VMware continues its ascendancy in the EMC universe. Last year's EMC World saw VMware given some prominence — almost like a gifted child allowed to perform for its elders, but not yet invited to the party. This year, Paul Maritz, CEO of VMware was given prominence above all other EMC execs — with the exception of Tucci. Indeed, both addressed the press together, both fronted the press Q and A. Tucci even commented that of all the brands that fall under the EMC umbrella, VMware was likely to be the only one that survived.

So what is this all about? EMC appears to be wholeheartedly embracing the vision of the virtual data center and the virtual infrastructure. As such, VMware moves front and center as a key element of the EMC hardware arsenal.

This can also be viewed in terms of the long-term steps EMC has been taking to ease of its dependency on hardware. At the start of the decade, the company was all about disk arrays. And then it suddenly embarked upon a stream of acquisitions as it rolled out its notion of information lifecycle management. The result is that while hardware takes up the bulk of EMC's $15 billion annual revenues, services and software now exceed one third.

"The storage hardware market doesn't look very pretty, with a 6.5 percent decline in the last quarter year over year," said Benjamin Woo, a storage systems researcher at IDC ( Framingham, Mass.). "The storage software and services sectors are stronger than hardware."

And VMware is now solidly in the driving seat of that move further into the software/services ecosystem. EMC has been working closely with VMware to integrate vSphere into the woof and warp of its storage systems.

"vSphere provides a layer of virtualization management," said Maritz. "vSphere is all about scale and signals the fact that we have moved out of the individual hypervisor business."

He believes this effectively marries up the touted benefits of the cloud and the data center — adding security, flexibility and manageability while facilitating the virtual data center.

"vSphere is a new layer of software to bring cloud benefits to the data center," said Maritz. "This could be likened to a data center OS or cloud OS."

While vSphere shares many of the attributes of a regular OS, it has interfaces that go down into the underlying hardware and upward into the application layer to provide security and infrastructure management. Meanwhile, to convince the world to adopt this virtual reality, VMware has been addressing one of the big hang-ups of virtualization in heavy-duty environments — lack of IO performance. 1 CPU with 2 GHz should be able to provide 10 VMs each with 200 MHz. Yet it rarely works out that way. So VMware has been cutting down on the resource consumption per VM while beefing up their IOPS range.

"We have reached a point where we have been able to generate a million IOPS from a single VM using the Nehalem processor," said Maritz.

In addition, the company has figured out software-based fault tolerance that manages virtual CPU resources with automatic failover for any workload

"With vSPhere, a new processor is reassigned right away based on it having a light workload," said Maritz. "vSphere is also fully aware of the underlying storage and can do thin provisioning to provide dramatic savings in storage efficiency by only provisioning storage when it is really needed."

As such, a VM can have its information moved from one SAN to another while still running. This simplifies the process of migration, upgrades and maintenance. Maritz gave the example of Terremark Worldwide (Miami). It offers customers access to a virtual pool of compute capacity. This eliminates the need to overprovision storage and server hardware for every application based on being able to satisfy peaks loads. By purchasing access to a 1000 GHz pool, for instance, you pay only for what you need.

"We made virtualization so easy that we created virtual machine sprawl," said Maritz. "Now, we are building a giant software mainframe that makes it possible for IT to offer a catalog of services where the user can define different levels of performance, related costs, and add such items as archiving, compliance and security for specific resources."

Cisco, too, is embracing this VMware dream. Its Nexus 100V, for instance, can be hooked directly into vSphere.

"We are reworking our management tools so OEMs like Cisco and service providers can embed them into their offerings to make federation more seamless." said Maritz.

This is all part of an ongoing initiative named Project Zoka, which is a joint development between EMC's Cloud Division and VMware. Its intent is to move the most valuable and complex applications onto the cloud along with their compliance and SLA requirements to make the virtual infrastructure more intelligent and aware.

This, in turn, drives a sea change in endpoint provisioning. Instead of the software being installed on a device, it can be installed via VMs to people. Wherever the person goes, that specific environment will be made available via VMware View. Software gets installed via templates for each user or type of user, i.e. install a Windows OS once into a template and that goes out to all users.

According to Maritz, this functionality will be available as part of vSphere later this year. Further, a vCenter plug-in for storage will be released during the second half of this year.

This article was originally published on Friday May 29th 2009
Mobile Site | Full Site