HP’s ‘shorty’ blade server takes fresh approach

Is HP’s BladeSystem c3000 Enclosure a blade server, or is it a modular, server put into a 6U rack profile? It’s a bit of both.

With its brand-new c3000 hybrid chassis, HP has remade its now-famous tower enclosure and the server’s guts into a flexible blade-enclosure format, retaining all the niceties of discrete servers but adding the flexibility of rack/blade modularity. HP’s c3000 Enclosure has a horizontal blade design that can accommodate as many as four full-width c-class blade devices or eight half-width server or storage blades.

Unlike HP’s higher-end c7000-class blades, the c3000-family blade server is not the typical blade enclosure designed to be piled high and deep inside a network operations center. Instead, the c3000 we tested seems best suited for branch offices where it’ll take up just the first six rack spaces, ostensibly sharing the rack real estate with other supporting equipment (routers, storage-area-network blocks and other network devices or appliances).

Archive of Network World tests

Subscribe to the Network Product Test Results newsletter

The overall performance of these blades was quite good, but we have to note that HP knew our blade server tests incorporate a “green” element — we measure the electricity required to drive these things — and shipped low-end CPUs, thereby optimizing it for low power consumption.

HP also supplied two server blades, the HP BL460c (based on a 64-bit Intel dual-core 1.6GHz Xeon 5110 CPU; this is the slowest one shipped on the blade); and the HP BL465c that uses Advanced Micro Devices’ 2110 HE CPU (1.8GHz, 64-bit, dual-core, also the slowest and smallest shipped with an AMD CPU). Both server blades came with 1GB of memory and have similar serial-attached-SCSI drive connections.

The performance characteristics of these slower server blades matched their clock rates. The slightly faster BL465c turned in a faster time of 392.3 microseconds vs. the BL460c’s 425.2 microsec in the LMBench3 (our usual comparison test) processor fork+execve test, which tests the speed of context shifting and memory movement and allocation.

The interesting part is that the AMD-based blade used 62 watts at peak during the test sequence, whereas the Intel blade needed 77 watts at peak. This compares with a recent test we did on IBM’s blades, which at a faster clock rate (2.66GHz) executed more quickly at 289.9 microsec in the processor fork+execve test, but used 89 watts at peak during the same test, with the same operating system — Fedora Core Linux — in the same configuration (albeit a quad-core CPU, rather than a dual-core). These numbers are somewhat hard to compare, but give an indication of good performance and power consumption.

The c3000 blade frame and its components are fully modular and built to order. As the options for components are staggering, HP has a Web site that lets customers select components via a Java/flash application. It’s a highly visual site, and specifics of each blade are selected by part type (computer blade, storage, tape and others) and by placing the selected components visually into the enclosure. HP should be commended for this new way of buying and configuring servers.

Storage options include drive arrays and tape drives. The drive arrays can be connected through the chassis or used as iSCSI targets from other internal or external host servers. Because the enclosure is designed to be a self-contained, mini-data center, the storage options — often missing from more data-center-focused blade servers — will be welcomed by small businesses and branches that don’t have access to high-speed, over-the-net backup architectures.

The HP StorageWorks Ultrium 440c tape blade stores data on LTO-2 cartridges, in increments of data starting at 400GB. We verified that 2:1 compression is standard. We also tested HP’s StorageWorks SB600c All-in-One storage blade, which combines a supplied Windows Server Standard Edition to create network-attached-storage shares. The storage also can be used as an iSCSI target for virtualized storage for the computer blades. While seemingly targeted for Microsoft NT File System storage needs, we also could use other iSCSI initiators to populate the server after a few configuration steps were found.

The SB600c blade was fast. Same-server read/writes (a file copy of a 2GB file) went at 340M bytes/sec (with no other process pending and in a minimum configuration in terms of services). Writes from the BL460c blade were slower, 146M bytes/sec of the same 2GB file. Then we tried the iSCSI target software and found that file copies through a blade iSCSI initiator went faster (which surprised us, because there should have been some overhead or driver-efficiency attenuation to slow things down) at 161M bytes/sec, using the same 2GB file. The iSCSI performance was impressive.

Several applications are supplied that together let IT managers service, support, administer, configure and monitor the c3000 system tested. At the heart of the enclosure is the Onboard Administrator, which also launches the integrated Lights Out 2 (iLO2) blade-management connection to each blade device, which can in turn, launch the administrator.

Onboard Administrator lives as an autonomous process application inside the system board of the c-class enclosure. Through it, we could view many characteristics of the enclosure, including the c-class system’s front-panel LED setup (which displays a health summary, basic IP settings, information regarding the enclosure and blade IDs, power consumption for the six fans on the enclosure, and PIN information, which is handy in branches or potentially physically unsecure areas). 

Unfortunately, the PIN can be as short as a single character, or there can be none at all — a bit of a security risk. Beyond the LED front-panel emulation, Onboard Administrator monitors each model’s power, heat and other characteristics, giving a basic health synopsis and reports and logs with e-mail triggers for the gradients (from critical to informational) of system problems.

ILO2 comes in two versions. The standard edition is bundled with the system, and there’s an advanced Blade Center Select license, which we received a demonstration of but did not test. The basic iLO2 functions are strongly focused on specific blade-device characteristics, and iLO2 observations are made via direct remote connections to agent software running on target blade devices.

The iLO2 controls include using a local, viewing machine’s media as virtual media for the iLO2-managed device you are connected to (handy for updates and installations), viewing power settings, administering security accessibility (including managing SSL certificates), and configuring network and SNMP settings.

Blade-server configuration wizards do such things as set RAID levels for blades, set up iLO2 agents on blades and then kick-start them (with a reboot). The optional Blade Center Select iLO2 license enables such things as two-factor authentication, group accounts, virtual-machine provisioning and setting higher or more thorough controls on power consumption. Some iLO2 functions require Firefox, and some of the selections for Systems Insight Manager (described next) require Internet Explorer.

There also is a 30-day teaseware license for HP’s Systems Insight Manager. This application has lots of niceties, such as managing virtual-machine deployments with precision (especially with VMware’s software) or monitoring a variety of operating system environments: Windows (NT through present), Linux (Novell’s SUSE and Red Hat) and the SCO Group’s UnixWare. Insight Manager deserves a review of its own, but it’s optional here and mentioned for those companies that would need its conveniences when rolling out multiple quantities of c3000. Oddly, some parts of Insight Manager require Internet Explorer 6.0+ explicitly. We used Firefox and found that Konqueror (KDE) and Safari (Apple’s browser) can give Onboard Administrator and iLO2 fits where it refuses to work.

The Onboard Administrator and iLO2 management and monitoring settings were good enough to pass our three “smoke tests.” We placed a plastic bag over the server to see if it would shut down without frying, because its settings were made to trigger shutdowns when the unit overheated. It took longer than we expected to fail in this test because power supplies would fail (there were also six fans), chill, then come alive again, redundantly. Eventually, the unit hit the “I’m going to cook” threshold and shut itself down, properly logging everything in detail as things failed and went into shutdown sequences. Just before and after restart, our e-mail box was flooded with warnings.

In our second smoke test, we “went rogue” and installed several operating system instances on the blades without telling Onboard Administrator or iLO2 to see if either could detect the changes. Because blade settings’ probes are scheduled, these applications took about a day to discover something was awry.

Once we also placed appropriate agents onto the blades, Onboard Administrator and iLO2 were very happy. We also found that even though a blade was misidentified, we still could access it remotely (the blade thought there was a Windows server on the blade, but we’d installed Fedora Core Linux) through the administrator’s Java KVM capabilities. We were impressed.

Finally, we pulled blades, simulated failures and generally mistreated the c3000, giving it the full user experience in terms of the strange things that users do. We received no unexpected failures (all server blades and modules restarted with operating-system complaints as expected, but all recovered within the constraints of the operating system), and we filled the logs with warnings, errors and reported jitteriness — completely and correctly.

We found the c3000 to be well designed for the Fortune 500,000 in terms of flexible component options, serviceability (blade changeout, and ease and understandability of hardware configurations) and management of the overall enclosure and its components. The management tools are strong, understandable, largely secure and capable. 

What we’d like to see are fewer management and monitoring software administrative overlaps, perhaps a convenient combination of iLO2 (with options) and Onboard Administrator, were it possible. Extended, this dream also would add the Systems Insight Manager with its options, unifying everything into a less overlapping, comprehensive, one-stop tool shop for using the BladeSystem c3000. It would make the difference between cold calculation against HP’s formidable competition and a truly compelling value deal.

Henderson is principal researcher and Dvorak is a researcher for ExtremeLabs in Indianapolis. Henderson can be reached at thenderson@extremelabs.com.

Henderson is also a member of the Network World Lab Alliance, a cooperative of the premier reviewers in the network industry, each bringing to bear years of practical experience on every review. For more Lab Alliance information, including what it takes to become a member, go to www.networkworld.com/alliance.