With virtualization's popularity soaring, it was predictable that hardware vendors would eventually bring to market specialized servers that cater to the needs of virtual machines. The market leader in this category is Dell, which currently offers three models of virtualization-optimized systems: the entry-level R805 server, and the larger R900 and R905 servers. Although these systems make perfectly good generic servers for all standard IT uses, they have specific features that endear them to virtualization users.
First, the machines provide extensively scalable RAM. One of the most pressing constraints to consolidating systems on a virtualization platform is RAM. This is due in part to the tendency of hypervisors to allocate the VM's full configuration of RAM, even if little of it is actually used; having lots of RAM is a boon to host systems. (And likewise, configuring VMs to have the minimum amount of RAM they need is a good practice.)
Second, the servers offer substantial network I/O. An important limitation of VM hosts is the amount of disk I/O they can support. Two VMs with heavy disk I/O can essentially freeze out the other VMs if the disk storage is local. For this reason (and others), VM hosts generally don't rely on local storage but have big pipes pointing to spindle farms. To get good performance, the connection is via Gigabit Ethernet or Fibre Channel; a good host provides many adapters for network I/O.
Finally, the systems have a means of embedding the VMware ESXi virtualization software directly in the hardware, which I'll discuss later. (The embedded virtualization can be disabled and another vendor's hypervisor layer, such as products from Citrix or Microsoft, can be used instead.) This embedded option makes it easy to configure the servers for virtualization right out of the box without the usual fuss of setting up the hypervisor and the attendant utilities. In addition, in the event of hard-disk failure, the virtualization software is still available separately.
As I mentioned earlier, Dell combines these features into systems aimed at sites that use virtualization, whereas other enterprise vendors such as Hewlett-Packard put these features to varying extents into various models in their product line and deem them all "virtualization ready."
I examined two virtualization-oriented servers from Dell, the R805 and R905 systems, which use AMD Opteron processors. (In Dell marketing numerology, servers and workstations ending in "05" use AMD processors, while those ending in "00" use Intel chips.) They deliver very good performance as well as the vaunted expandability and substantial networking I/O that VM hosts so badly need.
A peek inside
The systems are based on AMD Opteron processors running at 2.5GHz. The AMD processors, curiously, run consistently better on virtualization benchmarks than their Intel counterparts. The difference is small (a few percentage points) but consistent. Although the exact reasons are unclear, Dell attributes the boost to better performance of the floating-point operations used in most benchmarks. However, I suspect that the superior AMD memory management is a greater contributor. We'll know more on this point when we see the results of benchmarking VMs on Intel's upcoming Nehalem chips, which use the QuickPath Interconnect, Intel's equivalent of AMD's memory design.
The R805 has two Opteron processors, while the R905 has four. The systems I reviewed have 16GB and 32GB of RAM, respectively. They have 16 and 32 DIMM slots (respectively) for 667MHz DDR 2 RAM (PC 5300), which today have a ceiling of 8GB capacity per DIMM. So, total memory currently is 128GB and 256GB, respectively -- plenty of RAM to run lots of VMs.
The servers come with up to four Gigabit Ethernet adapters. Network adapters with 10GbE capacity are expected to ship before year end. The servers also accommodate Fibre Channel HBAs, which fit into the PCIe slots. The R805 can accept four of these adapters, while the R905 will handle a maximum of seven.
A final important feature, which supports Dell's description of these systems being virtualization-optimized, is the presence of the embedded VMware ESXi installation I mentioned earlier. This feature relies on a clever design. The VMware software is installed on a memory card that is socketed in the chassis. To boot from this card, an admin enters the server's setup menu and the card is enabled as a boot device. Then at boot time, the admin lists it as the default boot drive, and Bob's your uncle.
Both server models have this option. They can also install VMware the old-fashioned way, via a DVD, as they are both endowed with optical drives.
As mentioned earlier, most virtualization hosts do not rely extensively on local disk drives for storage due to the distinct possibility that a few VMs will overwhelm the drives' I/O bandwidth. However, these servers ship with drives that can contain ESX (if you don't use the embedded version) and easily hold the VM images that you run regularly. Options are traditional 3.5-inch disks or the emerging enterprise-speed 2.5-inch disks, both of which can be configured for RAID, using the built-in controller.
Benchmarking virtualization hosts
Although virtualization has been around for decades, its surge in adoption is a comparatively recent phenomenon. As such, it does not yet benefit from industry-standard benchmarks. The first widely accepted benchmark, called GrandSlam and designed by IBM, has been retired. A second suite designed by Intel, vConsolidate, runs database, Java, mail, and Web servers and computes a performance rating by combining their results. The general industry perception is that while the vConsolidate approach is valid, its specific implementation tends to disfavor AMD processors. So, it has quietly been left aside by the industry.
The best tests currently available are in a suite called VMmark from VMware (available at no cost from VMware's Web site). It is the most widely quoted benchmark currently in use. However, it is difficult to run (making it hard for in-house analysts to duplicate test results), and it tends to under-represent the importance of RAM. Hence, it's viewed as a useful measure, albeit one that's not truly representative of the typical IT profile. VMmark scores are measured in a peculiar unit called a "tile." As defined by VMware, a tile is a unit of work that aggregates different workloads running simultaneously on a system. The more tiles the system can run, the greater the system capacity. As the tile score computation contains a performance factor, it is fair to view tiles as a measure that represents both performance and scalability.
Examining the posted VMmark scores, we see that the R805 comes in at 7.96 tiles, and the R905 at 14.28. When compared with systems from other vendors, notably HP and IBM, these scores put the R905 at the top of the list in the category of 16-core servers, and the R805 in the middle of the pack.
Energy efficiency is a separate measure that is starting to see the emergence of vendor-neutral benchmarks, most notably SPECpower_ssj2008. This series of tests runs server-side Java (the "ssj" in the benchmark name) on the SUT (system under test) and determines a maximum workload. It then tests the power consumed at every 10 percent of the workload, takes an average of these, and publishes a single-number score that is the average number of ssj operations per watt of energy consumed. This number is possibly useful when comparing two servers in the abstract. But in the day-to-day work of an IT site, the number is problematic. What most sites want to know is how much work the server can do and how many watts it consumes. VMmark provides the former. And we'll now examine the latter.
The design of SPECpower as a ratio makes it comparable to the secondary information on the grocery shelves that tells you how much an ounce of cereal costs, but says nothing about the cost of the entire box or the quality of its contents. And even then, it's difficult to know how ssj and benchmark results map to your particular server. The ssj code is not Java EE-based, and it performs no database access, so the server-side tests are unlikely to duplicate activity at most IT sites. Moreover, the results assume a usage profile that operates the machine for equal periods at 10 percent, 20 percent, 30 percent of load -- all the way up to 100 percent. Again, it's not clear how this maps to actual server usage at most sites. So, for the nonce, I am sticking to the VMmark for performance/workload capacity and raw measures of watts at the wall for power consumed. (I measured the watts with the Kill a Watt meter, which is an excellent, inexpensive tool for measuring power usage.)
The R805 consumes 207 watts at rest (that is, running the operating system, but no virtual machines) and 411 watts under maximum load (running Dell's stress test software utilities). The R905 tips in at 331 watts at rest and 652 watts at full load, suggesting that performance (tiles per watt) is close to equal for both systems. The consumption numbers are good. The R805 barely consumes more power than a high-end workstation.
Head of the class
Dell's decision to provide virtualization-optimized servers is a smart move. Currently, it is the only server vendor to offer these systems. HP, by comparison, states that all its servers are optimized for virtualization, although this claim seems hard to accept in view of conflicting demands that might disfavor virtualization-friendly features. I suspect Dell's innovation will eventually require matching responses by both HP and IBM to cater to enterprise adoption of virtualization, especially in light of Microsoft's recently released hypervisor, which will further stimulate adoption.
The two machines are both very good solutions, the R805 being a journeyman and the R905 the greater professional. To my view, the R805 is a little limited and only SMB-class businesses will find its two-processor approach sufficient. Moreover, if a business wants local storage on its virtualization server, the two-disk ceiling will quickly prove limiting. Pricing and performance are good but not exceptional on this model.
The R905 is a far more complete solution. With double the number of processors, double the RAM capacity, and more than triple the disk space, it is a true enterprise or medium-sized business solution that provides plenty of room to grow. It also has the best posted performance for servers in its category. Given that, at retail, it's only US$5,600 more than the R805, it is easily the preferred solution and an excellent system.