Article

VMware overhauls ESX hypervisor

Alex Barrett, News Director
On Monday, VMware Inc. of Palo Alto, Calif., officially announced the upcoming release of its VMware Infrastructure suite.

    Requires Free Membership to View

    Login

For more on VMware's ESX and VI3 announcements:
VMware's new Infrastructure 3 offers features plus

VMware's new VI3 rejiggers bundles and pricing

The three components to VMware's release are enhancements to VMware's core ESX hypervisor platform with version 3.5. VMware has also added features and functionality in its Virtual Infrastructure suite; and rejiggered its pricing and bundles.

Over the past couple of months, many of the features included in the ESX 3.5 beta were leaked onto the Internet; this announcement confirms the vast majority of them. What is notable, however, is the fact that the new features and capabilities apply equally to traditional ESX with its service console, and to the new embedded ESX 3i, announced last month at VMworld. In addition, VMware announced standalone pricing for ESX 3i for customers that choose not to purchase servers from vendors like IBM Corp., Dell Inc., and Hewlett-Packard Co.

ESX 3i's smaller surface area is touted as being more secure than the full version, but it also has practical benefits: It's one less thing to patch.
,

At least one beta tester, Miami-based managed hosting provider Terremark Worldwide Inc., is leaning heavily toward the embedded version of ESX for future deployments. Jason Lochhead, principal architect, began deploying a beta version of the embedded hypervisor in August and found it much easier to install and manage. "One of the things is that we don't have to put disks in the [server], and we don't have to carve out a LUN [logical unit number] on the SAN [storage area network] for the boot image," Lochhead said. ESX 3i's smaller surface area is touted as being more secure than the full version, but it also has practical benefits, Lochhead said: It's one less thing to patch.

Ideally, Lochhead would like to buy ESX 3i from Terremark's preferred server vendor, HP. "I'd just as soon have it just show up." Barring that, Lochhead said building a bootable USB drive with the ESX 3i image was pretty simple too.

A hyped-up hypervisor
The core ESX 3.5 code brings many long-awaited features to the table, delivering new storage, networking, memory, processor and operating system support.

New storage capabilities. On the storage front, ESX 3.5 now supports lower-cost Serial ATA (SATA) drives for local disk instead of only high-performance SCSI and Serial-Attached SCSI. VMware shops investigating alternatives to Fibre Channel SANs will also welcome the addition of native support for 10 Gbit Ethernet and InfiniBand, while existing Fibre Channel SAN shops can now take advantage of N_Port ID Virtualization (NPIV), which assigns a virtual World Wide Name (VWWN) to each virtual machine associated with a Fibre Channel host bus adapter.

While NPIV support is not explicitly mentioned in the press release, "there is a subset of customers for whom NPIV is important," said Jon Bock, VMware senior product marketing manager. With it enabled, shops can opt to use Fibre Channel zoning techniques to expose LUNs on a SAN to individual virtual machines rather than to entire ESX hosts or clusters of hosts. "With NPIV, there's much more granularity in how you carve up this access," Bock said.

NPIV also promises to be a boon to storage resource management (SRM) vendors, because it gives them better visibility into individual virtual machines (VMs), said Mark Bowker, analyst at Enterprise Strategy Group (ESG) in Milford, Mass.

But in its first release, NPIV support is available only to users whose VMs reside on raw devices, not on VMFS virtual disks, Bock said. Plans are under way to support NPIV on virtual disks, but for the time being, "NPIV on raw devices was what was possible for us to release."

Improved CPU utilization. New networking capabilities in ESX 3.5 include support for Jumbo Frames, which allows Ethernet to transport larger frames for improved CPU utilization and increased throughput. Also included is support for TCP Segment Offload (TSO), in which part of the TCP processing is offloaded from the main CPU to a TSO-enabled Network Interface Card (NIC).

As far as memory and processor support are concerned, ESX 3.5 will now support the extended memory management features resident in the latest x86 microprocessors, namely Nested Page Tables on the Opteron processor from American Micro Devices Inc. (AMD) as well as Extended Page Tables on the Intel Virtualization Technology, or Intel VT, platform. "Until now, ESX only supported hardware assist for CPU virtualization," said Bogomil Balkansky, VMware's senior director of product marketing. Furthermore, ESX 3.5 also supports more memory per guest -- 64 GB, up from 16 GB -- and per physical host: 128 GB, up from 64 GB.

Finally, ESX 3.5 adds formal support for Linux guests that support the paravirt-ops standard. Those guests must run on Linux kernels 2.6.21 or later. To date, Ubuntu Linux 7.04.is the only shipping commercial Linux distribution to support the standard.

Accelerated deployment
For his part, Terremark's Lochhead expects to move to 3.5 relatively quickly -- within a couple of months of general availability, he said. "When we migrated from [ESX] 2.5 to 3.0, we had to do a VMFS migration, and that was pretty time-consuming," he said. Since moving to from 3.0 to 3.5 requires no such changes, "I suspect the migration will go pretty quick in this case."

Nor does Lochhead see anything on the competitive virtualization landscape that would derail Terremark's loyalty to VMware. "I keep my eyes out, but feature-wise, we haven't found anything that really meets our needs," he said. "I suspect we'll stick with VMware for the foreseeable future."

Let us know what you think about the story; email Alex Barrett, News Director.

.

Join the conversationComment

Share
Comments

    Results

    Contribute to the conversation

    All fields are required. Comments will appear at the bottom of the article.
    Back to top