Data management paper

Four-socket congurations can
accommodate up to 6 terabytes of
RAM and 60 cores, while eight-socket
congurations can provide up to 12
terabytes of RAM. More memory
provides greater analytical exibility by
allowing applications such as Terracotta
BigMemory Max to store and access
larger in-memory datasets. In a clustered
environment, Terracotta BigMemory
Max can scale out in-memory access to
petabytes of data.
Performance
Built on Java, Terracotta BigMemory
Max can run as a standalone application
using only the Java Development Kit
(JDK), or on many leading Java application
stacks, including IBM WebSphere*,
Oracle WebLogic*, and Apache Tomcat*.
Intel works closely with Java Virtual
Machine (JVM) vendors to increase the
performance of their JVMs on Intel
hardware. Applications such as BigMemory
Max benet from JVM optimization,
which takes advantage of the latest Intel
microarchitecture enhancements.
The Intel Xeon processor E7 v2 family also
provides a number of key performance
enhancements over the previous generation
of the Intel Xeon processor E7 family:
• Utilizes a 22 nm process technology,
which allows higher operating speeds
while reducing power consumption
and heat
• Allows for up to 15 cores and 30 logical
processors per socket—a 50 percent
increase over the previous generation
Intel Xeon processor E7 family
Reliability
The Intel Xeon processor E7 v2 family
provides enterprises with world-
class reliability and uptime. Previous
generations of Intel Xeon processors
introduced reliability, availability, and
serviceability (RAS) features that give
servers RISC-like capabilities. The Intel
Xeon processor E7 v2 family continues to
build upon these features to provide even
greater uptime and data integrity.
Designed for systems with 99.999%
uptime requirements, the Intel Xeon
processor E7 v2 family provides
continuous self-monitoring and self-
healing capabilities that rival those of
RISC-based systems.
3
Some of these features include:
• Machine Check Architecture (MCA)
Recovery, which lets CPUs and
operating systems compartmentalize
errors that could crash the server, such
as unrecoverable memory errors.
• MCA input/output (I/O), which provides
information on uncorrectable I/O errors
to the operating system. The operating
system or monitoring tools can then use
this information to determine the cause
of system errors and enable preventive
maintenance.
• MCA Recovery Execution Path, which
handles uncorrectable data errors
passed to the CPU. This feature
enables operating systems and
applications to assist in recovering
from errors that cannot be corrected
at the hardware level.
• Enhanced Machine Check Architecture
(eMCA) Gen 1, which provides enhanced
logging information to the operating
system and applications, which can use
this information to better diagnose
errors and proactively predict failures.
• Peripheral Component Interconnect
Express (PCIe) Live Error Recovery
(LER),
4
which provides recovery from
and containment of PCIe errors.
Lower Operating Costs
In recent benchmark tests using common
Apache Hadoop workloads, servers
equipped with the Intel Xeon processor E7
v2 family outperformed servers equipped
with the previous generation Intel Xeon
processor E5 family by up to 3.5 times.
5
With the Intel Xeon processor E7 v2 family,
enterprises can achieve the same or
greater benets by scaling their analytics
clusters up instead of out by running fewer,
higher-powered servers. An infrastructure
that follows the scale-up model can reduce
complexity and lower power, management,
and cooling costs, all while still providing
the computational benets of a larger
cluster running on lower-powered servers.
These traits can result in a lower overall
total cost of ownership (TCO).
The Performance Advantage of
Intel Technologies
One of the most important aspects of
high-performance applications is the
Figure 2. Terracotta BigMemory Max* can pool RAM from multiple servers in an array conguration,
and provides high availability and resiliency by mirroring the data across two or more servers in an
array conguration
BigMemory
Max
BigMemory
Max
BigMemory
Max
BigMemory Max Server Array
Terracotta BigMemory Max
*
Server Array Configuration
Scale Out
Scale Up
BigMemory
Max
Java
*
Application
BigMemory
Max
Java Application
Intel-Based
Mirror Server
Intel-Based
Active Server
Stripe
Figure 2: Terracotta BigMemory Max* can pool RAM from multiple servers
into an array configuration, and it protects data by mirroring the data across
two or more servers
4
Terracotta and Intel: Breaking Down Barriers to In-memory Big Data Management