Technologies Paper

With memory congurations up to 12
terabytes of RAM, enterprises can
increase the amount of in-memory
buffer used by MapReduce tasks, which
can signicantly reduce data movement
between memory and disk-based
storage. In servers with smaller memory
congurations, MapReduce tasks can
quickly ll the buffer, which means
that the contents of the buffer are
automatically moved—or “spilled”—out
of memory and onto disk. Moving data
from memory to disk is time-consuming
and also increases Java garbage collection
tasks, which negatively impacts node
performance. With a large memory server
conguration, systems engineers can
congure the buffer to use more available
memory to avoid spilling to disk altogether,
which reduces I/O and unnecessary Java
garbage collection. The Apache Hive Join
workload benchmark demonstrates the
benets of a larger memory conguration
by yielding a 29 percent performance
gain between the 512 GB and 1.5 TB
memory congurations, with an
estimated cost of only 20 percent for
the additional memory.
1
Enterprises can also realize lower power
and cooling costs by using servers
equipped with the Intel Xeon processor
E7 v2 family. On the CPU-intensive
K-means workload, the Intel Xeon
processor E7-4890 v2 consumed
approximately 1,162 watts of power,
while the previous generation Intel Xeon
processor E5 consumed approximately
466 watts of power. However, the
previous generation Intel Xeon processor
E5 took 3.5 times longer to complete the
same workload, which resulted in the
workload consuming a total of 1631
watts of power—29 percent more than
the Intel Xeon processor E7 v2 when
normalized to performance.
1
Apache Hadoop
*
Workload Performance
350
300
250
200
150
100
50
0
Figure 3: Servers equipped with the Intel Xeon E7-4890 v2 provide higher Apache Hadoop workload performance over servers
equipped with the the previous generation Intel Xeon E5-2680.
1
Sort
100%
262%
244%
TeraSort
292%
261%
100%
K-means
350%
354%
100%
Apache
*
Hive Join
320%
248%
100%
Page Rank
274%
275%
100%
Intel® Xeon® Processor E5-2680 with 128 GB RAM
Intel® Xeon® Processor E7-4890 v2 with 1.5 TB RAM Intel Xeon Processor E7-4890 v2 with 512 GB RAM
10 to 3
Average Workload
Performance Ratio
5
Accelerate Big Data Analysis with Intel® Technologies