Two Enterprise Blade Servers

Our test lab has been sounding like a busy airport for
three months now, ever since we brought in and set up three blade systems for
review. It has been an exciting three months, we must say, to get our hands
dirty on some of the most high-end equipment that goes into data centers (and to
hear some of our co-workers complain that they can't hear themselves think due
to the noise made by our airplanes...oops...blades!). A blade system isn't
really a single product, but more like a mini-data center with lots of servers,
networking devices, storage, etc all fitted into a small box. It's a different
matter of course that such a small box itself weighs more than half a tonne!

A blade system is an engineering marvel, with so much
compute capability built into such a small form factor. It's ideal for
organizations looking to do server consolidation to save precious real-estate in
their data centers. It can also be used for HPC (High-Performance Computing)
applications, or even to replace the entire server room of a small organization,
who could start with a few blade servers and then add more as their requirements
grow.

The three blade systems we evaluated were from Dell, NetWeb,
and Fujitsu. The first two have been reviewed in this issue, and the one from
Fujitsu will be covered in the next issue, because we couldn't finish testing it
completely, despite all the late nights and weekends we spent in office
evaluating all the blades. And the fact that the Fujitsu system was taken back
for a customer demo didn't help either. Hence, you'll see its review next month,
as we try and work out something with the folks at Fujitsu in the meantime to
carry out the remaining tests.

So here is a comprehensive analysis of the two blade
systems that we reviewed this time from Dell and NetWeb.

Dell PowerEdge M1000e Blade System

First up on the block was Dell's eleventh generation
PowerEdge M1000e blade enclosure. Up to 16 half-height or 8 full-height blades
can fit into its 10U chassis. Alternately, you could also put combinations of
half and full-height blades into the enclosure, and Dell offers three different
blade configurations for each type, meaning a total of six different blade
models available for you to choose from. The choice of CPUs for these blades
range from Intel's Xeon 5200, 5400, or 5500 series, or the powerful six-core
Opterons from AMD. The blades models follow a particular naming pattern. Model
numbers that end with a '0', viz M600, 610, etc are based on Intel based CPUs,
while those ending with a '5' are Opteron based, viz M605, 805, etc.

We received two full-height and one half-height blade for
review, with impressive configurations. One of the full-height blades sported
four quad-core Xeon X5570 CPUs running at 2.93 GHz each. It has a whopping 18
DIMM slots for DD3 memory, and the blade that shipped to us had 144 GB RAM,
fitted as 18 x 8 GB DDR3 modules. For storage, it contains four bays for taking
2.5 inch SAS drives. Ours came with 4 x 146 GB drives, each running at 10K rpm.

The second full-height blade had four six-core AMD Opteron
8439 CPUs running at 2.8 GHz each. This one supports even more memory than the
Intel-based full-height blade, with a whopping 24 DIMM slots. Ours came with 10
of them populated with 8 GB DDR2 modules. Storage capacity however, gets limited
in this blade, with only 2 drive bays, which can take up to 300 GB drives each.
For such a powerful machine, it would be nice to have four drive bays.

Rear of the Dell enclosure with redundant PSUs, cooling fans, CMC and the network fabric.	The Dell blades have a tool free design. Opening and physical inspection, addition of hardware to the blades is fairly straightforward.

The half-height blade servers are a little less powerful
than the full-height ones. Ours came with two quad-core Xeon 5570 processors, 96
GB RAM (8 GB x 12 DIMMs), and two 146 GB, 2.5 inch SAS drives.

Another small, but useful thing you'll find on these blades
is an LED indicator, which glows orange when there's a problem and the blade is
not functioning properly, else it glows a cool shade of blue. So at one glance,
a data center admin can tell which blades are working and which ones are down.

The chassis alone for the Dell machine weighs about 50 kg
and a fully populated unit with all the blades, switches, PSUs, cooling fans,
etc weighs about 182 kg! Considering that it's just a 10U unit, imagine the load
on the floor when you fit three of these in a 42U rack! So plan your data center
design accordingly if you're planning to install these babies.

Ease of Setup
Setting up the M1000E is a breeze. Simply insert all the modules you
require, power it on, connect a keyboard, display, and mouse, and power it on.
After that, the moment you plug in a blade server, the chassis automatically
recognizes the same. The setup program asks for drivers to install, which come
on a DVD along with the system. Post this, you can see the system booting, and
you can start loading an OS on it. Each blade takes around 2 minutes to
initialize and be ready. After that, the blade server's BIOS is displayed, post
which the OS takes another 2 minutes to boot up.

The Mid-Plane
As the name suggests, the mid-plane sits in the middle of the blade
enclosure and allows the blade servers to be connected into it from the front,
and the power distribution, network switches, and cooling fans to be plugged in
from the rear. It's essentially a printed circuit board with female connectors
for the blade servers. The good thing about the mid-plane is that it's scalable,
in that it's been designed to support 10 GbE & QDR Infiniband. This essentially
protects your investment so that future blade servers that come with these
technologies can be plugged right in and the mid-plane would happily support the
higher I/O bandwidth.

On-board LCD of the Dell blade server provides you basic networking & health info on the enclosure and the blades.

Since the blades are designed to draw cool air from the
front and throw the exhaust air from the rear, the midplane also houses
ventilation holes to allow the hot air to move out from the enclosure's rear.

Hot Swappability and redundancyJust about everything that
plugs into the Dell's chassis is hot swappable and redundant for
high-availability, be it the six PSUs (power supply units), the six I/O module
bays, the CMCs (Chassis Management Controller), the 9 cooling fans, or even the
blade servers themselves. We managed to plug in and pull out all components at
will, and the machine kept running 'without batting an LED!'.

The funny thing we found about the blade enclosure was its
redundancy modes. There's a 0 redundancy mode, which requires at least 3 PSUs to
be plugged in, without which the system won't run. With 2.3 kW per PSU, that's
nearly 7 kW of power with 3 PSUs plugged in. Surely if you initially require
very few components, say a single blade server, one network or pass-through
switch, one CMC, etc, then you wouldn't require so much spare power. The best
thing would have been was to allow for two PSUs, with 1+1 redundancy, so that if
one fails, the other takes over. Why should the buyer pay for the additional PSU?
The other redundant power modes supported by the m1000E are 3+1, and 3+3, where
the former implies that there's one spare PSU available on standby, while the
latter mode implies three PSUs on stand-by.

The power management feature of the web based console shows you the power consumption patterns of the individual server modules.

The six I/O modules bays support 3+3 redundancy, meaning
whichever three modules you choose can be backed up three redundant modules.
Likewise, there's an optional redundant module for the CMC.

The Network fabric
The number of I/O options provided in the M1000e are very impressive. Each
blade server supports two LAN ports on the motherboard by default, and you can
optionally add two mezzanine cards into each blade server, so that the system
can connect to multiple network topologies, be it a LAN, SAN, or Interprocess
Communication. The LAN ports can be connected through the mid-plane to either an
Ethernet pass-through switch, or an actual GbE switch, which could be either a
Dell PowerConnect or one of Cisco's Catalyst series of switches. The other two
mezzanine cards could either connect to 10 GbE modules or a Brocade Fiber
Channel switch.

Management System
The M1000e has an impressive management system, which can be accessed physically
from the blade system itself, or remotely over the network using a web browser.

There's an analog integrated KVM switch from Avocent, with
USB ports and a D-sub for connecting a keyboard, mouse, external DVD drive, and
a monitor. The KVM provides these connectivity options from the front as well as
the rear of the blade enclosure. The KVM allows you to access all 16 blades that
might be inserted into the enclosure. The good thing here is that you can insert
an external DVD drive and load an OS or other applications to any of the blade
servers.

The front of the m1000E enclosure also contains a sleek LCD
panel with a small GUI to indicate the status of various blade servers that are
inserted into the enclosure. You can access the IP addresses of the enclosure as
well as the blade servers inserted in it from this panel. The LCD panel can help
you quickly perform troubleshooting tasks on the blades as well as the
enclosure.

The third way to manage the M1000E blade server is remotely
over the network, using the CMC (Chassis Management Controller). The CMC
contains two Ethernet ports, one of which can be used to connect the system to
your network so that you can access it remotely. The second one can be used to
connect a machine directly to the blade system to access the management
software. There's also a serial port, in case you'd like to access the CLI of
the CMC.

The Management software
The CMC software can be accessed over the network using a web browser, and
provides a host of management features. It lets you control all components in
the blade enclosure. It color codes all components to quickly identify whether
they're functional or not. You can drill down into any component from the
software to get their detailed specs. Besides this, one really useful feature we
found in the software is its power monitoring capability. From the software
interface itself, you can monitor how much actual power is each blade consuming,
what's the cumulative power drawn in kWh, and even the peak consumption for each
blade.

Power Consumption
After powering on just the Blade enclosure and all the rear components (CMC,
fans, PSUs, I/O modules, etc) plugged in, the M1000e consumed around 360 Watts
of power in stable state. Note that we hadn't plugged in any of the blade
servers till this point. After plugging in the blades, we could measure the
power consumption of each remotely using the CMC. We found that the full-height
blade consumed 280W of power in steady state, and then we plugged in two
half-height blades. These consumed 141 W and 125 W of power respectively.

Temperature Control
We loved the cooling fan technology incorporated into this blade unit. The
cooling fans auto-adjust their RPM to keep the unit cool. In fact, just to test
their effectiveness, we pulled out a few fans while the machine was running,
hoping that the temperature inside the enclosure would go up. We waited for a
while, and kept a constant vigil on the internal temperature, which incidentally
is another good feature provided by the CMC. The temperature indeed increased
after about ten minutes, and the remaining fans really picked up speed after
that. The fans continued to spin hard until the temperature came back down.
During the period where the fans were spinning faster, we noticed a jump in the
overall power consumption.

Bottomline: The new M1000e Blade system has an
impressive set of features and hardware configurations supported. Combine this
with the ease of setup and management, and you have a powerful solution in your
hands for server consolidation or HPC. The pricing would of course depend upon
the components you choose, considering that there are so many different modules
in it.