Nocona injects more power in Dell PowerEdge line

Dell’s new dual-processor PowerEdge 1850 and PowerEdge 2850 servers — the company’s first to use Nocona — represent a significant upgrade for x86-based servers, and they will prove to be tough competition for servers running alternative microprocessors such as AMD’s Opteron.

With the Nocona-family Xeons, Intel has raised the bar, and Dell’s first offerings in the space take advantage of the chip’s 64-bit architecture despite a lack of software available to make the most of it. Nevertheless, by delivering superior performance, these models represent worthy upgrades to Dell’s previous servers — the PowerEdge 1750 and PowerEdge 2650.

More than enough

Much ink has been spilled about Intel’s adoption of AMD’s 64-bit extensions to the x86 architecture. Although those new instruction set extensions — which AMD calls AMD64 and which Intel calls EM64T (Extended Memory 64-Bit Technology) — are important, they’re not at the core of Dell’s new servers because not much 64-bit software currently exists to take advantage of them. In fact, those extensions didn’t play any role in my tests, given that the PowerEdge 1850 and PowerEdge 2850 were delivered running the ordinary 32-bit version of Windows Server 2003.

Other changes to the Xeon processors — faster clock speeds, bigger cache, faster buses, and SSE3 (Streaming single instruction multiple data Extensions 3) — play a more central role in these Dell servers, as I expect they do in other Nocona-based offerings from other vendors. The new chip set that Intel introduced along with the Nocona chips is also its first server infrastructure to support PCI Express.

PCI Express, an enhancement to PCI-X, offers system manufacturers a point-to-point communication path for on-board devices and expansion cards instead of multiple shared buses. Think of PCI Express as similar to going from Ethernet hubs to Ethernet switches and you’ll get the idea; the benefit is that high-bandwidth communications between specific devices — such as memory and Gigabit Ethernet, or memory and graphics cards — won’t hog all the bus bandwidth. Dell has implemented PCI Express as an option for these servers but has not yet developed any compatible cards, so the company didn’t provide that option on the review servers.

Big brother

Dell sent me these new dual-processor, rack-mount enterprise servers a few weeks before Intel’s official rollout of the server version of the Nocona. Other than the superficial variances in form factor and related peripherals They’re essentially identical. The PowerEdge 1750 is a 1U server with two internal SCSI drives and two PCI slots; the PowerEdge 2850 is a 2U server with six internal drives and three PCI slots.

Very much a conventional 2U Xeon server, the PowerEdge 2850 is similar to other servers in Dell’s 2×50 line and includes the small front-panel LCD diagnostic screen that debuted with the PowerEdge 2450 some years back. Internally, the system I tested came with 3.6GHz Nocona Xeons and had 1MB of Level 2 cache. The server is capable of handling as many as 16GB of 400MHz DDR2 RAM; Dell provided 1GB for the test server.

In terms of storage, this server offers six Ultra320 SCSI drive bays; Dell equipped this test system with their PERC/4e RAID controller and three 36GB disks. There is also an internal floppy and DVD-ROM drive. Dell continues to get points, from me at least, for packing six drives into a 2U housing; several other vendors still don’t provide that level of storage density in this form factor.

As mentioned earlier, Dell will be offering PCI Express with these servers, but initially, the only option is for three 64-bit 133MHz PCI-X. (The PCI Express configuration will be with one x4 lane card, one x8 lane card, and one slot for a legacy 64-bit 100MHz PCI-X card.) Frankly, given that PCI Express is still for early adopters, Dell’s expansion offerings are quite acceptable. The server contains dual Gigabit Ethernet NICs on the motherboard.

The high-availability features also are quite acceptable. The server contains two hot-swappable power supplies, each rated at 350 watts and equipped with a cooling fan. There are four separate hot-swappable fans inside the cabinet — but considering that Dell recommends that the server be shut down and taken out of the rack before removing the cover, it’s unlikely that they’ll be hot-swapped in a production environment. During the tests, I removed one fan while the system was running, and the other fans sped up to compensate for this, while both the LCD panel and the OpenManage software signaled there was a hardware problem.

Little brother

Given that the two servers were nearly identical, even to the point of using the same firmware and device drivers, I spent most of my time with the PowerEdge 2850.

Both servers contain the same dual 3.6GHz Xeon processors and the same 1GB DDR2 RAM (expandable to 16GB). The PowerEdge 1850 contained two 36GB Ultra320 SCSI drives but was not equipped with a RAID controller. It also has the capability of supporting dual PCI Express slots but is currently equipped with one 64-bit 100MHz and one 64-bit 133MHz PCI-X slot. It also comes equipped with the dual, on-board Gigabit Ethernet adapters.

Finally, Dell gets big kudos for managing to squeeze two hot-swappable power supplies into the server. Note that competitor Sun Microsystems can’t manage a feat like this with its single, nonremovable supply in its Sun Fire V20z. With the hot-swappable power, Dell has eliminated one of the historical weaknesses of 1U pizza-box servers. Because it’s rare to find this feature on a 1U server, I’ve given the PowerEdge 1850 a higher score in this area.

Management port

Dell continues to do a good job with management. The servers contain a daughter card called DRAC 4/I (the fourth-generation Dell Remote Access Controller), which acts as the onboard management processor. The card includes its own Fast Ethernet controller for out-of-band management or via a serial port. Solid features of the DRAC system are a Java-based console redirector, which can ease controlling the server remotely. The remote console can watch the system booting and can monitor for BIOS errors, which is a real improvement over the previous generation. The DRAC card also handles SNMP traps and communication with external management software.

DRAC supports Telnet-based management but unfortunately doesn’t allow or require SSH-based communications, which is an encrypted Telnet. That would be a slight improvement for organizations that like to use command-line tools.

DRAC — and Dell’s companion OpenManage software — has evolved considerably throughout the years; the new version also adds drive virtualization, which makes it easier to install software remotely via the management workstation’s CD-ROM drive. Overall, it’s essentially on par with ILO (Integrated Lights-Out), the management card that Hewlett-Packard includes with many of its ProLiant servers and that is widely considered the best in its class.

More than a speed bump

There are two possible perspectives to take in looking at Dell’s newest two-way server. One is to simply consider it as an incremental upgrade to Dell’s existing 32-bit Xeon-based server line. The new generation offers an improvement in speed, thanks in part to an increase in clock speed from 3.6GHz to 3.8GHz, an increase to 1MB of Level 2 cache (some older Xeons offer as much as 2MB of cache, which may be better for some applications), and a faster FSB (front-side bus) of 800MHz, up from 533MHz. If that’s how you’re looking at servers, then the PowerEdge 1850 and PowerEdge 2850 are obvious, not-too-controversial drop-in replacements for their predecessors. You get more speed,  better management, and some nice high-availability features for a decent price.

The other way is to look at these servers as the vanguard of a new generation of systems based on 64-bit versions of Linux and Windows and on PCI Express. In that case, it’s too early to issue a verdict because neither Linux nor Windows is currently available for EM64T and Dell has not yet shipped its PCI Express boards. In that regard, these new PowerEdge systems are worthy of evaluation as part of a next-generation computing migration. But because they’re a bit ahead of their time, it’s better to wait for the rest of the ecosystem to catch up.

In this particular case, I’m looking at the new PowerEdge boxes as better 32-bit PCI-based servers. From that perspective, they certainly are.