Cisco Cisco UCS B440 M1 High-Performance Blade Server Livre blanc
3
Principled Technologies, Inc.: Cisco blade solution vs. HP blade solution: Full
enclosure power testing
enclosure power testing
During Prime95 testing, we also measured the power usage of the Cisco UCS 6120XP 20-Port Fabric
Interconnect. Figure 4 shows the projected difference in actual power usage between the Cisco and HP blade
solutions with the addition of Cisco UCS 6120XP 20-Port Fabric Interconnect power as we extrapolate for
additional blade servers. The Cisco UCS 6120XP 20-Port Fabric Interconnect is necessary for operation of Cisco
blades, but is provisioned to support as many as 20 Cisco blade enclosures. We assume that, for each watt a
data center expends on electricity for server power, it spends an additional watt on power to cool the server and
power auxiliary equipment. To account for this additional consumption (sometimes referred to as Power Usage
Effectiveness, or PUE), we multiplied the power consumption by a factor of two. Data centers may have different
proportions of power and cooling costs because cooling efficiency and technology, rack densities, and other
factors affect cooling costs. We compared two Cisco UCS chassis to one HP c7000 chassis in order to maintain
blade count parity in groups of 16.
Interconnect. Figure 4 shows the projected difference in actual power usage between the Cisco and HP blade
solutions with the addition of Cisco UCS 6120XP 20-Port Fabric Interconnect power as we extrapolate for
additional blade servers. The Cisco UCS 6120XP 20-Port Fabric Interconnect is necessary for operation of Cisco
blades, but is provisioned to support as many as 20 Cisco blade enclosures. We assume that, for each watt a
data center expends on electricity for server power, it spends an additional watt on power to cool the server and
power auxiliary equipment. To account for this additional consumption (sometimes referred to as Power Usage
Effectiveness, or PUE), we multiplied the power consumption by a factor of two. Data centers may have different
proportions of power and cooling costs because cooling efficiency and technology, rack densities, and other
factors affect cooling costs. We compared two Cisco UCS chassis to one HP c7000 chassis in order to maintain
blade count parity in groups of 16.
Starting with the minimum of 16 blades, the Cisco blade solution uses 272.6 W
less than the HP blade solution. This gap increased with each set of two Cisco UCS chassis in our projections to
an 11,406.1W advantage for Cisco at higher counts of chassis and blades.
an 11,406.1W advantage for Cisco at higher counts of chassis and blades.
0
2,000
4,000
6,000
8,000
10,000
12,000
16
32
48
64
80
96
112
128
144
160
W
a
tt
s
Number of blades
Projected cumulative power savings using Cisco instead of HP
Figure 4: Projected aggregate savings in total power consumption (including cooling) using the Cisco blade
solution vs. the HP blade solution.
solution vs. the HP blade solution.
Using the data in Figure 4, we can calculate a different way to look at the power savings. By using 160 Cisco
blades instead of HP blades, you will save over 11 KW of power. This is nearly equivalent to the power of two HP
c7000 chassis or one chassis and its equivalent datacenter overhead. Once you reach 112 blades and beyond, it
is possible to fit an extra 10 percent more Cisco servers within a fixed datacenter power capacity.
blades instead of HP blades, you will save over 11 KW of power. This is nearly equivalent to the power of two HP
c7000 chassis or one chassis and its equivalent datacenter overhead. Once you reach 112 blades and beyond, it
is possible to fit an extra 10 percent more Cisco servers within a fixed datacenter power capacity.
Workload
Prime95 is Mersenne prime search software created by programmer George Woltman. This application runs in
the foreground, searching for a Mersenne prime number, which is almost 13 million digits long. We chose
Prime95 because it is a heavily CPU intensive ap
the foreground, searching for a Mersenne prime number, which is almost 13 million digits long. We chose
Prime95 because it is a heavily CPU intensive ap
plication, allowing us to measure the system’s power
consumption under full load. According to the Prime95 stress.txt readme file, “This program is a good stress test
for the CPU, memory, L1 and L2 caches, CPU cooling, and case cooling. The torture test runs continuously,
comparing your computer's results to results that are known to be correct.”
SPECjbb2005 is an industry-standard benchmark created by the Standard Performance Evaluation Corp. (SPEC)
to measure a server’s Java performance. (Note: SPEC and the SPECjbb2005 are trademarks of the Standard
Performance Evaluation Corporation.) SPEC modeled SPECjbb2005 on the three-tier client/server architecture,
with the middle layer as the primary focus. According to SPEC, “Random input selection represents the first (user)
for the CPU, memory, L1 and L2 caches, CPU cooling, and case cooling. The torture test runs continuously,
comparing your computer's results to results that are known to be correct.”
SPECjbb2005 is an industry-standard benchmark created by the Standard Performance Evaluation Corp. (SPEC)
to measure a server’s Java performance. (Note: SPEC and the SPECjbb2005 are trademarks of the Standard
Performance Evaluation Corporation.) SPEC modeled SPECjbb2005 on the three-tier client/server architecture,
with the middle layer as the primary focus. According to SPEC, “Random input selection represents the first (user)