Sun

Next Generation Systems Architectures A Sun Perspective

Franz Haberhauer

Technical Director Chief Technologist Client Solutions Germany [email protected]

Innovation Matters

$1.9B R&D

Sun's Three Franchises

Sun Microsystems A True Systems Company Developer Apps / Webservices Middleware Web

App

Dir

Msg

Msg

Operating System Server Network

Storage

...

The New Software Platform Shrink-Wrap App

Application scales from 1 to perhaps 100's of users

O/S Processor OS400 AS400

DGUX 88K

VMS VAX

Aegis 68K

Irix

MIPS

370

Solaris

Alpha

TM

SPARC®

Linux Ultrix

For Desktop: 1 copy SW = 1 copy HW

MVS

x86

AIX

Power

Server HP-UX Windows x86

Precision

The New Software Platform Comp

Shrink-Wrap

Comp

App

Comp

O/S

MW

Processor OS400 AS400

DGUX 88K

VMS VAX

Aegis 68K

Irix

MIPS

Alpha

MW

Networked Components

Comp MW

MVS 370

Solaris

TM

SPARC®

Linux Ultrix

MW

x86

AIX

Power

Server HP-UX Windows x86

Precision

Service must scale to 10,000's or millions of simultaneous users!

The New Software Platform Comp Comp

App

Comp

O/S

MW

MW

MW Comp MW

Processor OS400 AS400

DGUX 88K

VMS VAX

Aegis 68K

Irix

MIPS

MVS Solaris

Alpha

TM

TM

SPARC®

Linux Ultrix

The Java Platform

370

x86

AIX

Power

Server HP-UX Windows x86

Precision

XML The Windows Platform

Application Architecture Evolution

Services = Graphs of Interacting Components Client/Server

Objects

Web Application

Web Service

XDR CORBA RPC

RMI

COM+

XML

Next

Next Next

Directories Security Models Administration

XML

XML

Services Graphs

XML EJB App SQL DB

JSP Web LDAP Dir

cache/ filter

X 106 MSG

SLA Capability

Capacity

Computing Pools

Connectivity

Internet/ Intranet Storage Virtualization

Storage Pools

Midlets J2ME

New Challenges Virtualization ● Provisioning ● Infrastructure ● Services ● Telemetry ● Observability ●

Service Point Architecture SAN

SAN

NAS

NAS

Storage Network Presentation App

DB

Directory

Internet/ Intranet

Security

Policy

Management

Vertical Scaling / Scale Up / Data Facing

Multiprocessor Cluster /Blade Server Proc

Proc

Mem

Mem

I/O

Memory Switch

Cache coherant SMPs

●

Large main memory

●

One pool of processors ●

Cluster Management

Tightly coupled ●

High bandwidth, low latency

●

Flexible, dynamic scheduling by OS

●

Scaling within the node

●

Highly available nodes

●

Powerful I/O subsystems

Loosly coupled

●

Cheap, small nodes ●

I/O

●

●

Multiple OS Instances

1-4 way, few I/O slots

●

Scaling by the number of nodes

●

Load scheduling by a separate component (HW or SW) ●

Loadbalancer, Oracle RAC, App Server, Grid Job Queing System

●

Usually only when TA/Session/Job is startet

●

Highly parallel/parallelizable loads (Web, HPTC)

●

RAS through replication

Proc

Proc

Proc

Mem

Mem

Mem

I/O

I/O

I/O

Network Switch

Horizontal Scaling / Scale Out / Network Facing

Deutsch's 7 Fallacies of Networking ● ● ● ● ● ● ●

The Network is reliable Latency is zero Bandwidth is infinite The network is secure Topology doesn't change There is one administrator Transport cost is zero

Blade Shelf

A Horizontally Scaled System

Blade Server Evolution Platform Wave 1: Evolving thin server design to provide increased density, improved environmentals and better system management for enhanced throughput processing Platform Wave 2: Adopts the latest interconnect technology and optimizes the blade design to deliver superior transaction processing Evolution between Waves is achieved through enhanced service management

Market Adoption

Wave 2 Wave 1

2003

2005

A New Meaning of “System” What we did inside the E10K box…

●

●

We are doing to the network…

● ●

True scalability: Add performance without adding management complexity! “Soft configuration” and “Soft cabling” Multiple, secure domains But with a big difference: Heterogeneous elements Network becomes like SMP backplane

Trend: Traditional Tiers are Disaggregated and Recomposed Disaggregation of server state and function Computation, storage, network, software

Recomposition into optimized entities Appliances, intelligent storage, compute engines, EJB™ components, software services, load-balancers, etc.

Entities live in the network Collectively, they are the (new) computer

N1 Grid: The Computer that is the Network

Moore's Law

130 nm (100-400M) 90 nm (200-800M) 65 nm (300M-1B)

1T 1T

1B 1B 42M

1M 1M 4-bit 2.25K

32-bit 275K

64-bit 3.2M

Transistor density doubles every 18-24 months

8-bit 5K

1K

1

16-bit 29K

Planar transistor

1959

1970

1980

1990

2000

2010

2020

2030

UltraSPARC IV Die Data Instruction Cache Instruction Cache Data Cache Cache Instruction Logic Instruction Logic

ECC for L2 Tags

Core 0

DTLB

L2 Tags

FPU

DTLB Core 1

FPU

L2 Tags

ECC for L2 Tags

Challenges in Processor Design

Memory

Power

Complexity

Memory Bottleneck Relative Performance 10000 CPU Frequency DRAM Speeds 1000

2x Every 2 Years

Gap

100 10

2x Every 6 Years 1 1980

1985

1990

1995

2000

2005

Typical Complex High Frequency Processor Time Saved Thread

C

M

C

M

C

M Time

Memory Latency

Thread

C

M

C

M

Compute

C

M Time

Memory Latency

HURRY UP AND WAIT!

Compute

Note: Up to 75% Cycles Waiting for Memory

Throughput Computing ●

Basic idea –

●

Maximum application-level work performed for throughput

Exploit rich TLP in modern workloads – – –

Parallelism Pipeline simplicity Latency tolerance

Chip Multithreading (CMT) C

Thread 4

C

Thread 3

C

Thread 2 Thread 1

C

M M

M M

C C

C C

M M

M M

C C

C C

M M

M M Time

Memory Latency

Compute

CMT – A Simpler Design ●

Large caches

●

Superscalar design

●

Out-of-order execution

●

Very high clock rates

●

Deep pipelines

●

Speculative prefetches Simple Core: Step and Repeat

Faster time-to-market

Simpler by Design

Limited number of unique transistors

CMT—Multiple Multithreaded Cores Core 8 Core 7 Core 6 Core 5 Core 4 Core 3 Core 2 Core 1

Thread 4 Thread 3 Thread 2 Thread 1 Thread 4 Thread 3 Thread 2 Thread 1 Thread 4 Thread 3 Thread 2 Thread 1 Thread 4 Thread 3 Thread 2 Thread 1 Thread 4 Thread 3 Thread 2 Thread 1 Thread 4 Thread 3 Thread 2 Thread 1 Thread 4 Thread 3 Thread 2 Thread 1 Thread 4 Thread 3 Thread 2 Thread 1

Time Memory Latency

Compute

How Can CMT Deliver? Relative Performance Per Core

1.0

0.5

0.5X

10%

1X

100%

Core Size vs. Die Usage

How Can CMT Deliver?

100%

100% x 1 = 1x

10%

100%

50% x 10 = 5x

Throughput Networking: Changing the Rules Again Two tasks on two hardware threads Increase packet Hardware C C C C processing by devoting Thread 1 silicon and threads to Hardware Thread 2 network tasks Network Processing

Compute

Packet Processing

No missed packets

Context Switching/Interrupts

Chip Multithreading: Increase Network Efficiency

Niagara http://blogs.sun.com/roller/page/jonathan/20040910

That's what we call a system. A system built for internet workloads. Not for the expedience of a press release. ... Silicon for our Project Niagara chip: 8 cores * 4 threads per core = a 32-way computer. On a chip.

(And before you ask, yes, we are planning a nicer box when we ship :)

Throughput Gains: An Order of Magnitude Ahead SPARC: The Next Generation

SPARC Processor Roadmap Data Intensive

Rock

Today SPARC APL

UltraSPARC IV

UltraSPARC III 1050

1200

1X

UltraSPARC IV+

8X

1200

4X

2X

Niagara

Network Intensive UltraSPARC IIIi 1000

1Y

UltraSPARC IIIi+

1280

2Y

15Y

SPARC APL+

30X

Joining SPARC Forces Advanced Product Line (APL)

Ult ra

SPA R

C IV

Ul

Throu traSPARC gh IV+ Comp uting put Excel Design lence

Sun Fire

Thousands of Applications itical ge r c n o Missi ing Herita ut Comp

Fujitsu PRIMEPOWER S

C64 R A P

2004

64 ARC

V

V+

●

●

●

Optimized to address all network computing workloads Multiple product families (low, mid, high) Systems based on SPARC64 (jointly developed) and Niagara/Rock (Sun developed)

SP

2005

2006

Sun x86 System Roadmap

Beyond

Blade Systems 8 Socket Systems Nauticus Network Switch Sun W1100z W2100z Sun Fire V40z Sun Fire V20z Sun/AMD Alliance

Nov. 17, 2003

Coming Soon Q3 2004

July 30, 2004

July 30, 2004 Feb. 10, 2004

Coming Soon

AMD64 Benefits Besides 64 Bits ●

Full 32-bit compatibility –

●

Benefits: 32-bit Apps on 64-bit OS – – – – –

●

Both for i386 ABI and Linux More Memory - Full 4GB virtual adress space System Call and Library Optimizations PROT_EXEC Faster Kernel – read/write, mmap, etc. ● Large segmap, SEG_KPM, etc. Seamless operation

Apps ported to 64 bits can be faster – – – –

Twice # of Integer Registers Increased # of SSE Registers (128-bit register) Improved calling conventions PIC code no longer causes speed impact (Position Independent Code)

AMD Opteron & HyperTransport Advantages IA32 and IA64

AMD64

CPU

CPU

CPU

CPU

CPU

PCI-X bridge

PCI-X bridge

CPU

CPU

CPU

NorthBridge

Southbridge

●

Shared Frontside Bus between memory traffic and I/O is performance bottleneck

●

Modern, distributed architecture

Suns Multi-Platform OS-Strategy

●

SPARC and x86 maybe Power, Itanium 64 and 32-Bit –

● ●

Horizontal and vertical scaling

●

Sun and 3rd party hw

●

Established ISV acceptance

●

●

● ● ●

●

Runs Linux apps unchanged (x86) Defined release cycle Binary compatibility “write once – run forever” Directed innovation –

●

●

●

Complements Solaris on the x86 platform 32 and 64-Bit Horizontal scaling Sun and 3rd party hw – wide variety Starting ISV acceptance – Red Hat – SuSE Fokus on open source less binary compatibility – Solaris (UNIX) interoperability

●

Interoperability

Innovation in Operating Systems Solaris 10

Advanced Tracing

Predictive Self Healing Reduced System Downtime Automatic Service Restart

Extreme System Performance Fast TCP/IP Stack

N1 Grid Containers Software Partitioning

Live Monitoring of Production Systems

Trusted Solaris Military-Grade Security

Server Virtualization with N1 Grid Containers Network 192.9.9.1

www

192.9.9.2

192.9.9.3

store 192.9.9.4

appserver

B2B 192.9.9.5 oltp

Domain 1

Resource Management + Security Isolation Fault Isolation

Independent Users Separate Networks Independent Storage Isolated Containers

Disk Storage and File Systems

N1 Grid Containers Zones foo zone (foo.net)

beck zone (beck.org)

zone root: /aux0/blueslugs

zone root: /aux0/foonet

zone root: /aux0/beck

web services

login services

web services

enterprise services

network services

network services

core services

core services

core services

(BIND 9.2, sendmail)

zoneadmd

(inetd, ldap_cachemgr)

ce0:1

/usr

zcons

(ypbind, inetd, rpcbind)

zoneadmd

hme0:2

(BIND 8.3, sendmail)

hme0:1

zcons

/usr

/opt/yt

(ypbind, automountd)

(Apache 2.0)

zcons

(Oracle, Application Server)

(OpenSSH sshd 3.4)

/usr

(Apache 1.3.22, J2SE)

zoneadmd

zone management (zonecfg(1M), zoneadm(1M), zlogin(1), ...) core services

(inetd, rpcbind, ypbind, automountd, snmpd, dtlogin, sendmail, sshd, ...)

network device (hme0)

remote admin/monitoring platform administration (SNMP, SunMC, WBEM)

network device (ce0)

(syseventd, devfsadm, ...)

storage complex

Virtual Platform

blue zone (blueslugs.com)

Application Environment

(serviceprovider.com)

ce0:2

global zone

Virtual Machines versus Containers

HyperVisor Host OS

Hardware IBM LPAR HP VPAR EMC VMware

Container

Container

{

Container

Kernel

4000 tested on a V880

Kernel

One OS Instance

<1% Overhead

Kernel

OS

OS

{

OS

5-15%? Overhead

Multiple OS Instances

Solaris 10 OS Kernel

SPARC or x86

Server Virtualization

N1 Grid Containers - “N1 Grid In a Box” ●

●

●

Server Container 1

Container 2 Container 3

Container 4

Container 5

Solaris Domain 1

Solaris Domain 2

Dynamic Reconfiguration Dynamic System Domains N1 Grid Containers

N1 Grid – One System to Manage

The Datacenter as a Single System “n computers operating as 1”

Sun Microsystems A True Systems Company Developer Java Systems Apps / Webservices Middleware Web

App

Dir

Msg

Msg

Server Network

Storage

N1 Grid Operator

Operating System

...

Scaling the N1 Grid N1 Grid Distributed Datacenter

N1 Grid System

N1 Grid Server

N1 Grid System Workload Workflow, Monitoring & Policy Automation

N1 Grid Engine N1 Grid Service Provisioning System

Business Application Services N1 Grid Provisioning Server for Blades N1 Grid Containers N1 Grid Data Platform

System Infrastructure

N1 Grid Architectural Concepts ●

Service Virtualization ● ●

●

Efficient scaling of the number of instances ●

●

Separating the service from the platform "Movable" service components

Cloning reference installations

Service level management ● ●

Performance monitoring and modelling Capacity planning 48

There May Still Be A Lockin

Middleware App

Dir

Msg

Msg

Operating System Server Network

Storage

...

Binary Lockin

Web

Administrative Lockin

Apps / Webservices

Sun Software Product Stack Open Source Software Infrastructure Services Operating Systems Hardware

SUN JAVA SYSTEMS TM

Java Enterprise System

Java Desktop System

Java Mobility System

Java Card System

Java Studio N1 Grid Solaris SPARC

TM

Linux x86

Java Enterprise System TM

Q1 Directory Identity/Access TM

J2EE Application Web Portal E-Mail/Messaging Calendar Server Instant Messaging Collaboration Availability MPEG Streaming Grid Virtualization Solaris

Q2

Q3

Q4

Q1

Q2

Q3

Q4

S

Dev. Environment

Portal Server - Userman., Authentication, Authorization, SSO XML/JSP - Presentation, Transformation, Desktop, Peronalization - Secure Access (Search, Files, Mail, Applications, VPN...) Portlet Portlet Portlet Portlet Portlet Portlet - Device aware

SecureJava - Personalized ES Portal Server – Centralized + Access SRAP to+Services Mobile Access and Information + PKP

Access Manager & Directory Server Server

Java LDAP ES SSO UDDI Central Directory - User Profiles and - Passwords - Service Access Profiles ...Manager PKI

Web/App. Server

A

S S S SS S Web S S

Services

S

S

S CRM

S

S S Shop

Java ES Workflow Rules Identity Manager Adapters Adapters Adapters

S S SS SS

S S Server SAP Web Container EJB Container Java ESS WebWeb & Application S CMS S SServer S Portal

Web Server Web Container EJB Container

Identity Manager

Sync/Provisioning Engine

Java JavaES Studio IDE

Message Queue Message Oriented Middleware

Java ES Mess. Queue (JMS, SOAP, XML)

Comm. Collab. Server Comm. & & Collab. Server

Mail Server

Java ES Communication Calendar Server MTA and Mail Server Collaboration IM Server Calendar Server Services Address Book IM Server

Java ES Verfügbarkeit Cluster Cluster & & Grid Computing Performance Grid Engine

Java Enterprise System TM

Delivery Pricing Licensing

$100/Employee/Year

Summary ●

Sun's franchises –

●

Choice – –

●

Innovations matters and pays SPARC or x86-based systems Solaris or Linux on x86-based systems from Sun

Sun as a true systems vendor –

Pre-integrated Java Systems ●

– –

Licensing model fits to throughput computing and N1 Grid

Other components integratable N1 Grid to manage n systems as 1

Sun The Network is the Computer Franz Haberhauer [email protected]