While clustered or load-balanced servers can be an architecturally elegant way of scaling new applications, they present significant challenges for deployment and administration, ranging from ease of software deployment and monitoring to such mundane issues as cable management, power consumption, and heat dissipation. That’s where server blades, such as Hewlett-Packard’s ProLiant BL10e and BL20p Server Blades, fill an important niche while proving less expensive to buy, install, and manage than their conventional counterparts.
These two rack servers offer quite different approaches to the data centre. The ultra-high-density BL10e stuffs 20 tiny servers into a 3U-high rack space. Each BL10e blade is a slide-in card that contains one low-voltage 800MHz Intel Pentium III processor, one 2.5-inch, 40GB hard drive, up to 1GB RAM, and two Fast Ethernet network interfaces. There’s also an innovative adapter that can be plugged into the front of each blade to offer direct KVM (keyboard, video, and mouse) access. The trade-off: each BL10e server processor is slower than most modern laptops and has almost zero expandability.
The larger BL20p server blade uses a taller, wider, form factor to place eight separate blades in a 6U enclosure. They offer more processing power with dual 1.4GHz Pentium III processors. Of course, that’s still underpowered by modern standards; Compaq’s ProLiant DL360 has dual 2.8GHz Xeon processors in a 1U chassis. Power is provided by a separate 3U-high power enclosure, whose hot-swappable power supplies can accommodate two BL20p blade enclosures.
Those BL20p servers also offer two hot-swappable Ultra320 SCSI-based 36GB drives and up to 4GB RAM. Two Fast Ethernet interfaces, that can be upgraded to Gigabit Ethernet, are available for application use; a third is dedicated to Compaq’s ILO (Integrated Lights-Out) management processor, which provides remote administration and server console via an Ethernet connection and a Web browser.
While the hot-swappable SCSI drive capability adds a degree of high availability to the BL20p servers, especially when configured for RAID 1, these servers have limited expandability. There are no PCI slots, so forget adding Fibre Channel. And HP didn’t build in direct KVM access; the only way to access the server’s console is via the ILO.
HP shipped us both blade systems, plus their accompanying enclosures, power supplies, an Insight Manager Version 7.0 management station, and a RDP (Rapid Deployment Pack) server, preconfigured in a short 22U rack. These servers were preloaded with Red Hat Linux; HP also supports Windows 2000 Server.
The servers and enclosures were all cabled and pre-tested, with operating systems and Insight Manager installed and configured. As such, we weren’t able to assess the difficulty of setup, but we expect that physical installation would be easy due to the decreased cabling. However, software configuration and setup of the management structure would likely be more complex than for stand-alone servers, due in part to the addition of the blade enclosures as managed components.
Ongoing administration of full blade systems is done through HP’s Insight Manager. The BL10e requires an additional admin tool, which is launched via Insight Manager. Integration of those two tools is spotty and needs improvement.
The plethora of required passwords also poses a problem. An administrator must log in to each managed device to manage it, sometimes via a Web interface, sometimes via a Telnet window opened in Insight Manager. There is no federated identity management, so administrators must keep logging in over and over again, and there is no easy way to update admin passwords. This can cause security problems; administrators might be tempted to use a single name/password throughout the whole managed system and never change the passwords. HP should federate password management within Insight Manager, or tie it into an external scheme such as Microsoft’s Active Directory or Novell Directory Services.
The review setup also included HP’s RDP server, which captures and installs operating system and application images onto servers. While RDP is designed for HP’s entire server line, it lends itself well to the blade paradigm. We used RDP, for example, to assign a specific enclosure slot to a particular Web server image. When the server blade was removed and reinserted (to simulate swapping out defective hardware), RDP was able to reformat and reinstall the software fully automatically in about 40 minutes.
We were very impressed with the BL10e blade; its extreme density is ideal when many discrete servers must be installed into a tight space, such as for clustering, load balancing or hosting. It’s also worth considering for branch offices: a single 3U-high rack could serve as an entire datacenter, with e-mail, Web, file/print, directory services, firewall, and other servers in one box.
Meanwhile, the pedestrian BL20p is effectively a repackaged conventional 1U server with slower processors and limited expansion space. Applications for the BL20p would include those where its simplified cabling and quick-change capability compensates for its other limitations.
Alan Zeichick is a technology analyst specialising in networking and software development.