AdvancedTCA Makes the Most of Multicore Processors

The Advantages of Multicore
Multicore processors have become the norm. This is true largely
because, for the typical application, a processor spends about
75% of its time waiting for data. Decreases in the latency of
memory have not kept pace with increases in the speed of
processors. And at this point in time, adding extra cycles to the
processor doesn’t improve performance much. On top of which,
adding extra hertz to the clock increases power consumption.
Multicore processors, on the other hand, can produce higher
overall performance, even if the cores run a little more slowly.
(Assuming, of course, that the memory subsystem can handle
multiple parallel data accesses.) These performance increases,
naturally, are most apparent in applications that benefit from
parallel processing.

The Advantages of AdvancedTCA
AdvancedTCA (ATCA) is intended for applications that require
more processing power than a 19-inch rackmount server can
provide, and also for applications that require redundancy
and high reliability. In stark contrast to a “pizza box” server, an
ATCA chassis can support up to 14 blades connected across
the backplane with fabric switches. The chassis also provides
power and cooling.

The blades are connected across the ATCA backplane by a Fabric
(data) Interface as well as a Base (control) Interface. The data
plane is most commonly 10 Gbit Ethernet today, but is moving
quickly toward 40 Gbit Ethernet. The control plane uses 1 Gbit
Ethernet. Both interfaces can be made redundant, in which case
each blade is connected to both ATCA hubs/switches.
This arrangement makes ATCA systems more reliable and more
serviceable than external cabling. At the same time, isolating the
control and data planes improves system management capabilities
and preserves bandwidth for revenue-generating data plane
traffic. Traffic isolation is also a key system security feature
because it keeps the switch management interfaces beyond the
easy reach of data plane traffic.


ATCA for Compute-Intensive Applications
In compute-intensive applications, a large number of processors
must communicate with high throughput and very low
latency. In an ATCA chassis, Ethernet (10 Gbit and 40 Gbit) offers
the required data throughput, especially considering that some
Ethernet switches also support pass-through switching mode.
In pass-through switching, packet switching latency can be
lower than 500 ns because packet transmission starts before
the packet is fully received.
Interconnect redundancy can be easily traded for extra bandwidth.
It is possible to essentially double the ATCA system
bandwidth for compute-intensive applications which don’t
require the redundancy features of ATCA. This is done by using
the two hubs (Ethernet switches) in parallel.
For compute-intensive applications, ATCA also offers a tremendous
amount of computing density. For example, an
ATCA system with 14 of our A10200 blades featuring dual
Intel® 6-core “Westmere” processors yields 168 cores within
a single ATCA chassis, all interconnected via an in-chassis,
high-speed interconnect.
Compute-intensive applications usually need a lot of reliable,
fast storage capacity. ATCA addresses this need at the level of
the blade itself with local hard disks located on the blade or
on an RTM. In addition, dedicated ATCA storage blades can be
used within the system and accessed via high speed protocols.
Additionally, depending on the application needs, an external
storage array can be connected via Fibre Channel, FCoE, or iSCSI.

ATCA for Communication Applications
Communication applications demand high throughput as well as
efficient packet processing. They also usually benefit from parallel
processing, which is where multicore technology shines. And,
while multicore general purpose processors are excellent computing
devices, they are not usually able to efficiently get data in
and out at very high rates.


AdvancedTCA Makes the Most of Multicore Processors
For this purpose, packet processors are usually the best choice.
Packet processors are a distinct class of multicore processor,
purpose-built to efficiently move packetized data. Packet
processor blades such as our AT2-5800 with dual 16-core
OCTEON™ Plus processors are readily available in the ATCA
blade form factor, which allows system designers to take advantage
of both types of resources within the same system.

Ethernet switches (ATCA hubs) can add value within this type of
system because of their load distribution capabilities. The most
current Ethernet switches are able to steer packets to a specific
ATCA blade thanks to their Access Control List features. Such
policy-based routing allows packet streams to be distributed at
very high rates (10 Gbit/sec to 100 Gbit/sec) with certainty that
packets belonging to the same flow are always directed to the
same blade.
A high-performance ATCA communication system composed of
two Ethernet hubs, two of our A10200 processor blades, and up
to 12 of our dual 16-core OCTEON Plus packet processor blades
provides 320 MIPS64 cores (OCTEON devices) and 24 x86 cores
(Intel Westmere devices).
Data comes into the Ethernet hubs and is distributed among
the packet processor blades based on policies. Packets are
then distributed between two OCTEON devices on the blade,
and then amongst the cores of the OCTEON device. Most of
the high-throughput processing is done by the packet processors.
If specific packets require more extensive processing, they
are sent to the x86-based blades. This is an efficient strategy
because most packets require minimal processing, but a few
will need more attention.

ATCA Virtualization
In a virtualized environment, multiple physical servers can be
consolidated onto a single multicore processor. These virtual
servers run on the same processor, but have no relationship
with each other because they are running on separate cores.
Hardware, with the help of Hypervisor, ensures that each server’s
operating system can safely access its own memory and I/O
devices without disturbing its neighbors.
No matter what the form factor, virtualization is a great way to
take advantage of multiple processor cores. ATCA makes the
most of this concept because it allows multiple blades with multiple
multicore processors to be consolidated in a single chassis.
Indeed, entire racks of legacy servers can be reduced to a single
chassis. Virtualization in ATCA also offers redundancy, because
a high availability, virtualized operating system allows an application
to be migrated from one physical server to another if a
hardware failure occurs.

Multicore Development Tools
Packet processors were specifically designed for parallel processing,
and therefore they already have software development
tools available for application development in a multicore environment.
Cavium’s OCTEON and other similar devices are based
on standard processor architectures (MIPS64 in the case of
Cavium) which run standard operating systems such as Linux®.
Nevertheless, packet processors perform best when running
simplified proprietary operating systems, such as Cavium’s
Simple Executive. The packet processors currently in use are
programmed using C and C++, even when their proprietary OS
is being used. There are also several software toolsets available,
including 6WINDGate software, which seamlessly manage the
relationship between x86 processors and packet processors,
significantly simplifying software development efforts.

Conclusion
Multicore processors fit nicely into the ATCA infrastructure, and
deliver very high compute density, reliability and redundancy.
The redundant, high-speed ATCA backplane supports both
communication and compute-intensive applications. Ethernet
switches can provide load balancing and policy-based routing
to the blades in a chassis, which prevents bottlenecks and
makes the most of multicore devices. Virtualization makes it
possible for multiple applications run on the same processor on
an ATCA blade, which also takes full advantage of the available
cores. And mature software toolsets help take the sting out of
application development.

www.ge-ip.com