The VPOD: Breakthrough Operational Efficiency Improvement For Data Centers

SDN at Light Speed
Introducing the VPOD:
CALIENT Technologies, August 2015
Breakthrough Operational Efficiency
Improvement For Data Centers

2 © 2015 CALIENT Technologies – All Rights Reserved
We Improve Data Center Operational Efficiency
 Reduced CAPEX & Improved Energy Efficiency
• Compute resources can be reallocated and shared at the optical layer, supporting
much improved % utilization of server and storage assets. These represent 85% of
the DC electronics cost and almost 70% of the energy consumption.
 Reduced OPEX and Improved Customer Satisfaction
• Automation of Fiber Infrastructure speeds new service provisioning, improves
customer satisfaction and retention, supports creation of new on-demand services,
improves reliability, eliminates truck-rolls, and lowers workforce cost.
Enabling “Lights Out” Data Centers

Problem: Poor Compute Resource Utilization
The biggest operational efficiency problem in Data Centers
Amortized Equipment Cost 1 Power Consumption 2
Sources: (1) James Hamilton, Amazon. (2) The Datacenter as a Computer, Luiz André Barroso, Jimmy Clidaras, Urs Hölzle (Google)
Compute Resources = 85% of Electronics Cost & 68% of Energy
Problem: Compute Resources are typically less than 40% utilized
Massive CAPEX
& Energy Waste

Why Is Compute Resource Utilization Low?
 Large Data Centers are organized into Pods consisting of
multiple racks, and sharing a common spine switch.
 AKA “Stamps”, Clusters
 Favored for performance and maintenance reasons
 A large Data Center may have hundreds of PODS
 The compute resources within the POD are very tightly coupled
for optimum performance with very low latency connections.
 Data Center Managers prefer to keep applications within a
POD for optimum performance – Resources are typically not
shared between PODS
 However, compute resource demand requirements for specific
applications fluctuates considerably based on time of day,
instantaneous events, etc.
 As a result, PODs are designed for worst-case (maximum)
resource demand so they are considerably under-utilized for a
large percentage of the time
How can resources be shared between PODs
without sacrificing performance??

Introducing a Breakthrough Technology
CALIENT’s S-Series 3D-MEMS Optical Circuit Switch (OCS)
 Up to 320 User Ports – 640 Single Mode Fibers
• 320x320, 160x160 options
 10, 40, 100 Gbit/s per port and beyond
 <50ms Maximum Setup Time
Introduces The Ability to Optimize Resources at The Physical Layer
 Less than 50ns latency
 Less than 3.5 dB Insertion Loss
 Ultra low power (<45w), small size (7RU)
 TL1, SNMP, OpenFlow APIs
Pure Optical-Layer
Circuit Switching

Back To Our Problem….
How can we share resources between PODs
to improve server and storage % utilization
without sacrificing performance??

CALIENT’s LightConnect™ Switching Fabric
OCS-Enabled Virtual PODs (VPODs) Support Resource Sharing
 Any resource (Rack, Row, Row Group) can be directly connected to any spine switch at Layer 1 = VPOD!
 VPOD always has tightly coupled low-latency resources regardless of size
 PODs can expand and contract with compute demand by sharing resources
 VPODs can be provisioned manually or in response to DC resource management tools
Compute Resources are physically shared between PODs in response to Application/Workload demands
CALIENT LightConnect™
Optical Switching Fabric
LightConnect™ Optical Fabric ManagerData Center Resource Management Tools/Systems

The power of the VPOD architecture is that it is not necessary
to over-build every POD with compute resources to handle
worst case anticipated demands.
Instead, compute resources are shared and reallocated on
demand between PODs at the optical layer without any loss of
performance.
As a result, PODs can be operated at much greater capacity
and operational efficiency and this translates directly to lower
compute resource CAPEX and power consumption.

Economic Benefit of LightConnect™ Fabric
Data Center Organized into Dynamic VPODs
 A small increase in Server Utilization
translates into a huge savings in total
data center spending
 Model assumes $2B in mega-scale
data center spending
• 85% of total electronic CAPEX is on
server resources
• 15% of total electronic CAPEX is on
networking
 A server utilization increase of less
than 10% is equivalent to saving the
total CAPEX of the entire data center
network
Savings is in the $100Ms for a Mega-scale Data Center

LightConnect™ Fabric Manager
 Provides access to all of the OCS nodes that are part of
the flexible LightConnect Fabric topology.
• Each VPOD with associated spine switch ports, and each
compute resource unit with associated ToR or end-of-row
switch ports are defined in the LightConnect Fabric Manager
topology and cross-connect management modules.
• Resource moves are made simply by connecting the resource
to the appropriate VPOD with the Fabric Manager.
• In addition to pushing the OCS cross connect state to each of
the OCS nodes, the Fabric Manager can trigger the push of
configuration changes to the ToR, end-of-row, and spine
switches.
 Deployment Options:
• Can be deployed as a stand-alone solution, allowing data
center operators to make changes to the fabric manually
• Alternatively, the Fabric Manager can be integrated with
existing data center resource management systems via variety
of open interfaces including TL1, REST, SNMP, or RPC

LightConnect™ Fabric Manager
Software Architecture

Further Benefits of Optical Network Automation
 Accelerate New Service revenue
• Provision in seconds instead of days
• Create and reallocate capacity on demand
• Enhance customer satisfaction and retention
 Reduce Operational Expense
• Reduce on-facility workforce and truck-rolls
• Manage connections from a central offsite operations facility
• Support automated entry and updating of facility electronic records
 Improve Quality of Service
• Minimize human-error scenarios
• Accelerate fault detection and isolation during service degradations or outages

CALIENT Technologies at a Glance
 Headquarters in Santa Barbara, CA, U.S.A.
• Atiq Raza (Chairman and CEO) held former roles as CEO of RMI, President and COO of AMD, and CEO ofNexGen
• Dr. John Bowers (Co-Founder) is Director of the Institute for Energy Efficiency at University of California Santa Barbara (UCSB)
 Optical Circuit Switching market leader. >$400M invested to develop & perfect core 3D MEMS optical technology
 2013: Started shipping new S-Series Optical Circuit Switch in volume to one of the largest global cloud data
centers
• The only Optical Circuit Switch shipping in high volume and tested to rigorous Telcordia GR63 & GR1073 requirements
 Data Center applications focus on improving operational efficiency by increasing compute resource utilization
 High Growth: In 2015 CALIENT ships our entire 2012 volume every 2 weeks. Driven by Data Center shipments

Thank You

The VPOD: Breakthrough Operational Efficiency Improvement For Data Centers

More Related Content

What's hot (19)

Similar to The VPOD: Breakthrough Operational Efficiency Improvement For Data Centers (20)

Recently uploaded (20)

The VPOD: Breakthrough Operational Efficiency Improvement For Data Centers