Evolution of Server Scalability
Evolution of Server Scalability
Bladed Bladed and and Rack Rack serversservers
Multi-Socket / Multi-coreMulti-Socket / Multi-core CPUs
CPUs
X64 X64 platforms platforms (Intel / (Intel / AMD)AMD)
Commoditized OSCommoditized OS
Virtual Machine DensityVirtual Machine Density
Management complexity
Management complexity
Cloud Computing / Dynamic
Cloud Computing / Dynamic
Resourcing
Resourcing
Monolithic serversMonolithic servers
Large numbers of CPUsLarge numbers of CPUs
Proprietary platformProprietary platform
Proprietary OSProprietary OS
Many apps per serverMany apps per server
High cost / Proprietary
High cost / Proprietary
Large failure domain
Large failure domain
Scale Up
Scale Up
CommoditizCommoditized ed serversservers
1 APP / 1 Physical Server1 APP / 1 Physical Server
X86 platformX86 platform Commoditized OSCommoditized OS Servers under-utilized Servers under-utilized
Power & cooling
Power & cooling
Scale Out
Scale Out
Scale In
Scale In
The 9
Evolution of Server Scalability: Changes
in server design
Console, power,
networking, and
storage
connectivity to
each blade
Console, power,
networking, and
storage connectivity
shared in chassis
Evolution of Server Scalability: CPU
Density
core core
core core
single
core
CPU
single socket for 1 CPU single socket for 1 CPU single socket, 1 CPU, 4processing cores
• Sockets – Slot in machine board for processing chip • CPU – Processing Chip
• Core – the actual processing unit inside CPU
Terminology
Server Impact
• More cores in a CPU = More Processing
• Critical for application that become processing bound • Core densities are increasing 2/4/6/8/12/16
Evolution of Server Scalability: Memory
Density
DIMM Slots DIMM
• DIMMs – Dual Inline Memory Module - a series of dynamic
random-access memory integrated circuits. These modules are mounted on a printed circuit board
• Ranking – memory modules with 2 or more sets of DRAM chips
connected to the same address and data buses. Each such set is called a rank. 1 dual and quad ranks exist
• Speed – Measured in MHz most server memory is DDR3 and
PC3-10600 = 1333 MHz
• As Server memory increases clock speed will sometimes drop in
Adapter Buses - PCIe
PCIe BUS
In virtually all server compute platforms PCIe bus serves as the primary motherboard-level
interconnect to hardware
•
Interconnect:
A connection or link between 2 PCIe ports, can consist of 1or more lanes
•
Lanes:
A lane is composed of a transmit and receive pair of differentiallines. PCIe slots can have 1 to 32 lanes. Data transmits bi-directionally over lane pairs. PCIe x16 where x16 represents the # lanes the card can use
•
Form Factor:
A PCIe card fits into a slot of its physical size or largerGrowing Use of Platform Virtualization
Platform Virtualization:
• Physical servers host multiple Virtual Servers • Better physical server utilization (Using more of
existing resources)
• Virtual Servers are managed like physical • Access to physical resources on server are
shared
• Access to resources are controlled by
hypervisor on physical host
Key Technology for :
• VDI / VXI
• Server Consolidation • Cloud Services
• DR
Challenges:
• Pushing Complexity into virtualization
• Who manages what when everything is virtual • Integrated and virtualization aware products
Server Management Challenge
Server Mgr Network Mgr Chassis Mgr Server Mgr Network Mgr Chassis Mgr Server Mgr Network Mgr Chassis MgrServer Orchestrators / Manager of Manager
Characteristics of Movement Toward Cloud
1960
1970
1980
1990
2000
2010
Mainframe
Mini Comp
Client Srv
Web
Virtualization
Cloud
Define the Nature of Typical Cloud Services
Platform as a Service:
Providing data center infrastructure via the cloud
Cloud Computing
IaaS
Infrastructure as a Service
Providing infrastructure for cloud based services.
Software as a Service:
Providing software infrastructure via cloud
Management • 3Tera • RightScale • Scala • Vertabra Applications • SalesForce • Gmail • BaseCamp • Square Space Platforms: • Python – Google App Engine • Facebook • Appistry • Force.com Service Providers
• Amazon Web Services • Joyent
Cloud Layer Review
Compute
- UCS B Series - UCS C SeriesNetwork
- FCoE - Nexus 7K, 5K, 4K, 3K, 2K,Virtualization
- Nexus 1KV - VM-FEX - UCS - A-FEX - VIC - Virt Appliance - VSGOrchestration / Management / Monitoring
- Tidal, New Scale, Altiris
- UCSM ECO Partner Integration (MS, IBM, EMC, HP) - UCSM XML API
VDI
CRM
Web
Store
Service Catalog
Orchestration
and
Management
Infrastructure
Mgmt Server
Server Deployment Today
Over the past 20 years
•
An evolution of size, not thinking
•More servers & switches than ever
•Management applied, not integrated
•
Virtualization has amplified the problem
Result
•
More points of management
•
More difficult to maintain policy coherence
•More difficult to secure
Mgmt Server
Unified Computing Solution
Embed management
Unify fabrics
Optimize virtualization
Remove unnecessary
–switches,
–adapters,
–management modules
Less than 1/3rd the
support infrastructure for
a given workload
Unified Computing System (UCS)
Single Point of Management
Unified Fabric
UCS Manager Embedded
Manages entire system
UCS Fabric Interconnect
UCS Fabric Extender Remote line card
UCS Blade Server Chassis Flexible bay configurations
UCS Rack / Blade Server Industry-standard architecture
UCS Virtual Adapters Choice of multipleadapters
Cisco UCS and Nexus Technology
CNAs with FCoE Nexus 2000 Fabric Extender Nexus 5000/ 5500 Unified Fabric
Nexus Products
UCS Components
System Components - Logical
Fabric Interconnect
Chassis
–Up to 8 half width blades or 4 full width blades
Fabric Extender
–Host to uplink traffic engineering
Adapter
– Adapter for single OS and hypervisor systems
Compute Blade
- Half Width or Full Width
Compute Chassis x86 Computer x86 Computer X I I x8 x8 x8 x8 B MGMT S S B X X X X X C C A G G G G SAN G R A G G G G R G P M P SAN LAN Fabric Interconnect Fabric Interconnect Fabric Extender Fabric Extender Compute Blade (Half slot) Adapter Compute Blade (Full slot) Adapter Adapter
System Topology
Mgmt SAN A
LAN
SAN B
Embedded Management (UCS Manager)
Single point of device management
– Adapters, blades, chassis, LAN & SAN connectivity – Embedded manager
– GUI &CLI
Standard APIs for systems management
– XML, SMASH-CLP, WSMAN, IPMI, SNMP – SDK for commercial & custom implementations
Designed for multi-tenancy
– RBAC, organizations, pools & policies
UCS Manager XML API Custom Portal Standard APIs CLI GUI Systems Management Software
Hardware “State” Abstraction
Separate firmware, addresses, and parameter settings from server hardware Physical servers become interchangeable hardware components
Easy to move OS & applications across server hardware
State abstracted from hardware
LAN Connectivity OS & Application SAN Connectivity
Chassis-1/Blade-2 Chassis-8/Blade-5 LAN SAN UUID: 56 4dcd3f 59 5b… MAC : 08:00:69:02:01:FC WWN: 5080020000075740 Boot Order: SAN, LAN BMC Firmware MAC Address NIC Firmware NIC Settings Drive Controller F/W Drive Firmware UUID BIOS Firmware BIOS Settings Boot Order WWN Address HBA Firmware HBA Settings
Service Profiles
Contain server state information
User-defined
– Each profile can be individually created – Profiles can be generated from a template
Applied to physical blades at run time
– Without profiles, blades are just anonymous hardware
components
Consistent and simplified server deployment – “pay-as-you-grow” deployment
– Configure once, purchase & deploy on an “as-needed” basis
Simplified server upgrades – minimize risk
– Simply disassociate server profile from existing chassis/blade and associate to new chassis/blade
Enhanced server availability – purchase fewer servers for HA
– Use same pool of standby servers for multiple server types – simply apply appropriate profile during failover Run-time association Server Name UUID MAC WWN Boot info LAN Config SAN Config Server Name UUID MAC WWN Boot info LAN Config SAN Config Server Name UUID, MAC,WWN Boot info firmware
LAN, SAN Config Firmware…
