The new Sun Fire X2270 and Sun Fire X4270 servers are the fastest x64 servers Sun has ever produced.
Sun's fast and cheap Nehalem-powered Sun Fire X2270 and Sun Fire X4270 servers promise to put some sizzle into Oracle's hardware business.
The Sun X2270 is a low-cost, 1U rack mount system. The X4270 is the X2270's big brother, a 2U system. Both servers can run one or two Intel Nehalem CPUs, from the 2.0GHz E5504s to the high-end 2.93GHz X5570s. But whereas the X2270 packs a lot of compute power in a somewhat constrained chassis, the X4270 offers slightly more power in a much more expansive box. Our evaluation Sun Fire X2270 had two X5570 CPUs and 24GB of DDR3 RAM.
The quick skinny: the Sun Fire X2270 would do extremely well as a front-end Web server, a small database server, or a member of a virtualisation farm, with the addition of a few NICs or an HBA. It's constrained by a single power supply, a single PCIe slot, only a pair of NICs, and four disk drive bays, but the low cost offsets these limitations, depending on the application.
Sun Fire X2270: Virtual test bench
To test the Sun Fire X2270, we opted for our baseline VMware test application, which is a LAMP stack packaged as a vSphere vApp with four VMs. This test is designed to mimic a large, database-driven web application, using a randomised mix of dynamic and static page delivery.
It's built on four CentOS 5.3 servers: a single MySQL server built with four vCPUs and 8GB of RAM, two web front-end servers with two vCPUs and 4GB of RAM each, and a load balancer with a single vCPU and 1GB of RAM. The web servers run a tweaked Apache 2.2 web server, with content mounted on an NFS share to the database server. The database server runs a highly tweaked MySQL 5.1.25 installation and exports the Web root to the front-end servers. All load balancing is handled by Nginx, running in the load balancer VM.
The test is built with nine vCPUs on purpose, in order to eclipse the eight physical cores present in the servers under test. Also, the static/dynamic call ratio, though randomised, is seeded to bring all boxes to a maximum load equal to the number of vCPUs in each box. The VMs communicate across an internal vSwitch, with only the load balancer directly linking to the lab network. All load generation was driven from ab, the Apache benchmarking tool, running 100,000 requests per test pass, 20 concurrent connections.
The Sun Fire X2270 is big on CPU and RAM but short on most other assets. In keeping with the Nehalem design, it can address up to 96GB of DDR3 RAM across 12 DIMM slots. It has "just" four hot-swap 3.5-inch SATA drive bays up front, two Gigabit Ethernet ports rather than the "normal" four that most Sun servers can claim, and a single PCIe 2.0 16x low-profile expansion slot, all backed up by one 600W power supply. It does include the Sun ILOM for remote management, with full graphical support out of the box.
In the lab, the X2270 moved like a much more expensive system. We did two test runs: one with the vApp running first on a single 500GB SATA drive, then another with the VMs housed on an NFS share to a SAS array run from an Adaptec Snap Server 650. The difference was noticeable and resulted in a performance increase of around 15 percent. With the single local disk against a RAID 5 array of SAS drives on the filer, this isn't surprising. In fact, applications that are more disk I/O intensive should show an even greater performance increase.
There's lots of power in this little package. The only downsides are the two Gigabit Ethernet interfaces, rather than Sun's normal allotment of four, and the lack of a redundant power option. In many applications, a server like this will need more than two Ethernet interfaces, and redundant power is always a plus. But for raw cost/performance, the Sun Fire X2270 is a very good deal.