Small, Dense, and Very Sharp

When historians compile the great accomplishments of 2002, the popular “Ketchup Song,” from the Spanish trio Las Ketchup, will be number one on the list. High-density blade servers will follow a close second. Considered by enterprise and service-provider IT staff to be the ultimate in high-density computing, blade servers go a step beyond the former “space champion,” the dual-processor, 1U-high (1.75-inch) high-density server. While blades are expensive and somewhat proprietary, if space is at a premium, you’ll eat ‘em up by the six-pack. Or maybe by the 24-pack, as you’ll soon see.

A server On a board

The blade server concept is straightforward, and is ideal for Linux computing: start with a modern single-processor server with an embedded Ethernet controller, and gradually strip away everything that doesn’t contribute to its computational horsepower.

Throw away the cabinet — it’s just taking up space — leaving a bare motherboard with integrated processor, memory, hard-disk controllers, and Ethernet ports. Lose the power supply — you can feed the machine juice from an external source. Deep six the CD-ROM drive and boot floppy. Yank out the video chip, mouse, and keyboard connectors — instead, use telnet or SNMP to communicate with the server. Finally, add a couple of hard drives, and affix them directly to the motherboard.

What’s left is 100% pure computational muscle, a server-on-a-board. Put that board into a slide-in cartridge, and you have a server blade that can be installed edge-wise into a special rack-mounted enclosure. The enclosure provides the power and cooling, as well as connectivity to a network. The blade takes care of the computing.

So why are blades so new? Mostly because blades required additional elements, such as network-attached storage, to be successful. When enterprises began installing identical load-balanced servers as front-tier Web servers or middle-tier application servers, the demand for inexpensive carbon-copy servers increased. Finally, clusters and real-time, stateless failover within applications make each server a simple cog in a complex machine. With blades, the entire server itself is a hot-swap unit.

So, what’s not to like? Only two things.

First, blades are not interchangeable between different manufacturers’ enclosures. In fact, in some cases, as with Hewlett-Packard, blades are not even compatible within a single manufacturer’s different product families.

And the other downside? With some blades, the computing engine is the weakest part of the story. For example, the smallest blades use single 677 to 800 MHz Intel Pentium III processors, a couple of generations below the state-of-the-art.

Of course, if you’re running Linux, even a Pentium III processor makes a mighty powerful server engine. Moreover companies do offer Xeon-powered blade servers, or at least the option to run dual Pentium IIIs. If that’s not enough horsepower, you should probably look at a conventional server.

So who makes x86-based blade servers? You’ll find the usual suspects: Dell, HP, and IBM, though Dell is rather late to market in this category (the long-awaited Dell PowerEdge 1655MC server started shipping in December 2002). There are also some lesser-known players in the blade-server game. One that’s noteworthy in the Linux community is RLX, which shipped its first blade server in May 2001.

As you’ll see, each company’s blade products are unique, not only reflecting “state of the art” technology, but each brand’s vision of the future. For example, out of today’s available offerings, IBM offers the highest performance blades and Fibre Channel connectivity, while HP and RLX boast the highest server density.

Dell Dabbles in Blades

Dell (http://www.dell.com) is known for its broad support for selling fast computers at low prices. The company’s 1U-high dual-processor servers are among the most innovative rack-mount machines, often a full processor generation ahead of competitors. However, Dell’s PowerEdge 1655MC blade server, the company’s only offering in this class, is not particularly compelling. It’s not very dense, and it’s not very fast.

Figure One: The Dell PowerEdge 1655MC

The PowerEdge 1655MC, shown in Figure One, allows for six blades within a 3U-high (5.25-inch) enclosure. While that’s twice the density you’d achieve if you simply installed three of the company’s 1U-high servers, it only lets you install 84 servers in a standard 42U-high rack — not bad, but not close to what HP and RLX offer.

First, let’s look at the enclosure. Dell provides two redundant power supplies, each of which can serve the entire rack. Redundant power is critical because if you lose power to an enclosure, a whole set of servers goes off-line. The enclosure also includes an integrated remote management card with a built-in keyboard/video/mouse switch and a managed Gigabit Ethernet switch (that means you only need to have one Ethernet cable running to the blade enclosure). For redundancy, you can add a second Gigabit Ethernet switch — that switch has its own independent connections to the individual blades, so you have full failover there as well.

The proof of the pudding is in the eating, or in this case, the serving. Each of the six blades can be equipped with one or two 1.26 GHz Pentium III processors, and as much as 2 GB ECC SDRAM. Each blade contains a single-channel Ultra320 SCSI RAID controller, and there’s space to mount one or two SCSI hard drives, each of which may be up to 73 GB in size. That’s a speedy processor, but far short of the 2.4 GHz Xeon chips that IBM offers with its blades.

To install software on the server blade, there’s a single USB port which can be used with an external floppy or a bootable CD-ROM drive. You can also use the Ethernet connections to do remote installs.

However, there aren’t any PCI expansion slots, so you can’t connect the Dell blades to a Fiber Channel storage area network. Frankly, it’s hard to see much to commend in the Dell product.

Hewlett-Packard Offers Two of a Kind

Hewlett-Packard offers two separate and incompatible blade server products in its ProLiant product line (yes, both were former Compaq models). Information can be found at http://www.compaq.com/products/servers/platforms.

Figure Two: The HP BL20p

Figure Three: The HP BL10e

The BL20p, pictured in Figure Two, hosts eight high-performance dual-processor blades in a 6U (10.5 inches) high rack enclosure — that’s only 56 servers in a 42U-high tower rack. By contrast, the BL10e, shown in Figure Three, is a high-density system, cramming as many as twenty single-processor server blades into a 3U-high enclosure. That’s 280 servers in a rack, second only to RLX’s offering. Granted, those are relatively underpowered servers, but that’s just incredible server density.

Starting with the BL20p, the enclosure is designed for high availability, with two separate interconnects that keep all the servers plugged into the six hot-swappable power supplies. You can also optionally outfit each enclosure with one or two Fast Ethernet or Gigabit Ethernet switches.

And that’s about all you’ll get from the enclosure. HP doesn’t include the Ethernet or keyboard/video/mouse switches the way that Dell does. Instead, you talk to the servers over your network, using the specially integrated “Lights Out Management” chip on each server blade. The chip allows you to communicate with the blades using a Web browser, even if the server has crashed.

And about the blades: with the BL20p, you can have one or two 1.4 GHz Pentium III processors, similar to those on the Dell PowerEdge 1655MC. Memory is expandable to 4 GB. Each blade contains three Fast Ethernet ports, plus a fourth for the “lights out” management port. Two of those ports can be upgraded to Gigabit Ethernet. You won’t find a PCI expansion slot, or even a USB port. You’ll have to do all your troubleshooting over the network. However, each blade has an integrated Ultra160 SCSI controller (without RAID) that can work with two on-board and hot-swappable SCSI hard drives.

The high-density BP10e server is a horse of a different color. In the 3U-high enclosure you can have as many as twenty bare-bones blades, each with either a single 700 MHz or 800 MHz Pentium III processor. The enclosure contains two hot-plug power supplies and a single integrated Fast Ethernet switch (if it goes down, you are sunk, but then again, Ethernet switches are very reliable). There’s no KVM. All management is done over the network via Telnet or RS-232 serial connections.

Figure Four: An individual blade from the BL10e

The tiny blades themselves are impressive (shown in Figure Four). Although they don’t have the latest multi-gigahertz processors, they do allow as much as 1 GB of RAM, and have two integrated Fast Ethernet ports. There’s also an integrated ATA-100 disk controller. The 700 MHz version has a 30 MB 2.5-inch hard drive; the 800 MHz one comes with a 40 GB 2.5-inch disk. That’s right, a laptop hard drive. That’s one of the tricks the Compaq engineers used to achieve the high density. (RLX also uses laptop disks, but they’re slower ATA-66 drives.) Unfortunately, the Compaq engineers also left out all the “extraneous” pieces, like the USB port and expansion ports. But imagine the idea of collapsing a load-balanced Web farm, tied into a backend database, into a single, 3U-high, twenty server cluster using blades.

Big Blue Blades Blaze Brightly

IBM’s blade solution (http://www.pc.ibm.com/us/eserver/xseries) consists of separately named components. The large 7U-high enclosure, called BladeCenter, can handle as many as 14 dual-processor blades, yielding a maximum of 84 servers in a single standard 42U-high rack, the same as Dell’s offering.

Figure Five: The IBM HS20

The biggest difference is that IBM’s blade, called the HS20, is packed with power in the form of one or two Intel Xeon processors. IBM blades can also connect to Fiber Channel SANs directly through a plug-in adapter card, which greatly facilitates integration into a large, centrally managed storage pool.

IBM has added other innovative features into their BladeCenter enclosure, such as a floppy and CD-ROM drive that can be shared by the individual blade servers. The enclosure also has two hot-swap power supplies, but two additional ones can be added, and are recommended for when the box is fully laden. You can add as many as four Gigabit Ethernet switches right into the BladeCenter enclosure. There’s also space for adding an optional keyboard-video-mouse switch.

When it comes to the blades themselves, IBM has packed in a lot of power. Given the larger physical dimensions of the blades, there’s plenty of room to add in larger processors and support circuitry. The HS20 is the only blade here that offers one or two 2.0 GHz or 2.4 GHz Xeon processors, an advanced server processor based on the Pentium 4 chip design. They also allow a maximum of 4 GB RAM. There are also two Gigabit Ethernet controllers on each blade.

For storage, IBM includes an ATA-100 disk controller, and IBM currently offers as many as two 40 GB ATA-100 drives for the HS20 server. The company also offers an interesting twist: if you want Ultra320 SCSI storage, you can buy a SCSI Storage Expansion Unit, which attaches directly to a server blade and provides both the SCSI controller and space for two hot-swap drives. The down side is that this expansion unit doubles the size of the blade, and thus you can only install seven SCSI-equipped blades into the enclosure. You can also use IBM’s Fibre Channel expansion card to connect to an external SAN switch, and from there link each blade to an external storage unit.

Big Blue’s blades stand out. IBM offers a lot of flexibility with its servers, and processing power, too.

RLX Really Racks ‘em In

RLX Technologies (http://www.rlx.com) is a smaller hardware vendor that specializes in high-density servers. The company offers several different enclosures, as well as a set of different blades, exceeding the variety from any other vendor.

With RLX, let’s start with the blades themselves, and then move to enclosures. The blades come in three separate models: two small, one large, yet all are only single-processor cards, somewhat similar to Compaq’s BP10e blade.

The smaller blades, at 1.5-cm thick, are the ServerBlade 677 and ServerBlade 800i, based on the 677 MHz Transmeta Crusoe processor or 800 MHz Intel Pentium III processors (the i in the product name is for Intel). Both of these blades, which are physically interchangeable, include as much as 1 GB RAM, dual embedded Fast Ethernet controllers, and an embedded ATA-66 disk controller. Both cards can be equipped with one or two 2.5-inch 40 GB hard drives.

Figure Six: An individial RLX ServerBlade 1200i

The larger blade, which is 3.0-cm thick, is the ServerBlade 1200i. It includes a single 1.2 GHz Pentium III processor, and can hold 2 GB RAM. It has the same dual Fast Ethernet network ports and ATA-66 disk controller. This card can be populated with as many as two 40 GB or 60 GB hard drives;. The 60 GB drives are thicker, and can’t be used on the thinner cards.

Now, to the enclosures, which the company offers in two sizes. Unique to their product line is the System 100ex enclosure, a 1U (1.75-inch) rack-mounted pizza-box that can hold as many as six of the 1.5-cm blades — that is, the ServerBlade 667 or 800i. That lets you cram 252 servers into a standard rack. The System 100ex also includes two hot-plugging power supplies.

The bigger container is the System 300ex, which is a 3U-high box that can hold as many as 24 of the company’s 1.5-cm blades, increasing density to a whopping 336 servers in a standard 42U rack. The System 300ex can also be used to host twelve of the ServerBlade 1200i cards or a combination of the larger and smaller blades. It also has dual power supplies.

Beyond the high density, the RLX product line lacks many of the extras found within the other companies’ blade lines. There aren’t any dual-processor or Xeon-based servers, and there’s no support for Fibre Channel or Gigabit Ethernet. There’s not even a KVM switch, as you’ll find in the Dell solution, or a USB port for connecting floppy or CD-ROM drives.

However, when it comes to maximum server density, RLX currently stands alone.

Blades Almost As Cool As Ketchup

Blade servers certainly aren’t for the typical small or medium-sized business — or even for the typical large enterprise. While they certainly do boost the server density of a data center’s racks, they offer other compromises, including slower processors, limited expandability, and limited internal storage. They also generally lack keyboards and displays. While this isn’t a problem for a properly running Linux server, it could present a problem while building up a system or troubleshooting startup problems.

Are blades for you? If you have a need for many small servers — perhaps to run a Web cluster, or to improve the efficiency of a hosting environment — the blade concept may fit your needs perfectly. If your servers are running identical software, rely upon external storage or middleware links to other applications, and are front-ended by a network load balancer, you’d probably benefit from the blade concept of using individual servers as plug-and-play devices. Beyond that, say as a general replacement for a company’s stand-alone servers, the specific advantages of a blade may not have any value. Indeed, for transaction processing, email servers, and database back-ends, traditional multiprocessor servers are still the best way to go.

But you’ve got to admit that stuffing 20 and 24 servers into a 5.25-inch-high rack enclosure is pretty darn cool.

Almost as cool as Lola, Lucia and Pilar Muños, who single that delightful Ketchup Song. Jé ja!