thin clients: thin clients: back to the future back to the future - - PowerPoint PPT Presentation

thin clients: thin clients: back to the future back to the future

<Jason Nieh> nieh@cs.columbia.edu <Jason Nieh> nieh@cs.columbia.edu

“Computers in the future may weigh no more than 1.5 tons.” a Popular Mechanics editorial 1949

PCs in use worldwide (2004) PCs in use worldwide (2004)

224 69 53 46 36 29 26 23 22 19 19 14 13 11 11 206

5 0 5 0 100 100 150 150 200 200 250 250 U S U S Japan apan C h i n a n a G e r m any any U K U K F rance ance K ore a e a I ta l y C anada anada B ra z a z i l i l R u s s u s s i a A u s u st r t ra l i l i a I nd nd i a M e x e x i c o c o Ne th e h er l ands ands O th e h er C

• u

n t r y P C P Cs i n u s e u s e ( m illi illi

• n s
• n s )

"I think there is a world market for about five computers” remark attributed to Thomas J. Watson Chairman of the Board International Business Machines 1943

“There is no reason anyone would want a computer in their home.” Ken Olson founder and chairman Digital Equipment 1977

“I can assure you that data processing is a fad that won't last out the year.” a Prentice Hall business book editor 1957

today today’ ’s computer s computer

today today’ ’s computer problem s computer problem

problem #1: manageability problem #1: manageability

problem #2: operating cost problem #2: operating cost

move add change $1000 per

incident

problem #3: availability problem #3: availability

problem #4: work area problem #4: work area

problem #5: security problem #5: security

dis dis-integration of the computer

• integration of the computer

thin-client computing thin-client computing

network decouples computing and display

user input

thin-client computing thin-client computing

display updates application processing and data here stateless client secure server room

benefits benefits

 simplify IT management  minimize cost of desktop failures  transparent user mobility  continuous computing access  secure computing services  improve user/computer work areas  utilize resources efficiently

trends trends

computers are cheap people are expensive

computers computers vs vs people people

 Dell Dimension

2400 PC desktop, 2.4 GHz CPU, $420

 Dell PowerEdge

420 server, 2.4 GHz CPU, $350

 move, add,

change: $1000 per incident

“the network is the computer”

thin clients thin clients vs vs mainframes mainframes

 dumb plain text

terminals

 large,

expensive machine

 rich graphical

interfaces

 clusters of

inexpensive servers, blades

thin clients thin clients vs vs web web

 rewrite

applications for the web

 client browser

and helper applications

 preserve

software investments

 no client

applications

key technologies key technologies

remote display remote display

display pipeline display pipeline

applications window system display device driver framebuffer

client/server partitioning client/server partitioning

applications window system display device driver framebuffer client server

window server on client window server on client

applications window system display device driver framebuffer client server

window server on client window server on client

 local non-app UI interactions  complex software running on client  software needs to be maintained  client resources need to scale

window server on server window server on server

applications window system display device driver framebuffer client server

window server on server window server on server

 no complex client software  no client software maintenance  client scales with display  “ultra-thin” client

wire protocol wire protocol

 high-level graphics  low-level graphics  2D drawing primitives  raw pixels

application application   protocol protocol

applications window system device driver framebuffer raw pixels high-level requests

THINC THINC

applications window system device driver framebuffer raw pixels high-level requests

virtual device driver

THINC THINC

applications window system

display updates input events

THINC protocol THINC protocol

 copy  solid fill  pixmap fill  bitmap  RAW

THINC optimizations THINC optimizations

 offscreen drawing  transparent video support  local cursor drawing support  server-push model  smallest update first scheduling  server-side screen scaling

enables enables

 stateless clients  heterogeneous display devices  remote access  remote collaboration

performance? performance?

yes yes no yes yes yes no yes yes yes audio Linux no 24 yes no X Linux no 24 yes yes VNC Linux yes 24 yes yes Sun Ray Linux yes 24 yes no THINC Linux yes 24 no no NX Linux yes 24 no no X ssh -C Windows yes 8 yes no GoToMyPC Windows yes 24 yes no RDP Windows yes 24 yes no ICA Linux N/A 24 N/A no local PC OS en- crypt ion color depth ultra thin custom window server name

configurations configurations

 desktop LAN  desktop WAN  802.11g PDA

desktop LAN web performance desktop LAN web performance

0 . 7 0 . 59 0 . 53 0 . 5 0 . 56 0 . 53 0 . 44 1 . 01 0 . 43 N/A 0 . 5 1 1 . 5 2 2 . 5 3 P C P C ICA CA R D P R D P GoT o M yP C X X s s s sh

• C

N X N X VN C S u n S u n R ay TH IN C

p e p er w eb eb page page l a tency ency (s )

desktop LAN web performance desktop LAN web performance

25 . 29 147 . 32 203 . 91 391 . 8 108 . 4 59 . 33 205 . 54 228 . 92 126 . 75 N / A 5 0 5 0 1 0 0 1 0 0 1 5 0 1 5 0 2 0 0 2 0 0 2 5 0 2 5 0 3 0 0 3 0 0 3 5 0 3 5 0 4 0 0 4 0 0 P C P C I CA RD P G o G oToM yP C P C X X s s s sh - C N X N X VN VNC S u n S u n R a R ay TH I N C

p e p e r w e b e b page page d a d a t a ( K B K B )

desktop WAN web performance desktop WAN web performance

0 . 77 0 . 79 1 . 05 2 . 89 1 . 46 1 . 47 0 . 64 0 . 85 1 . 21 0 . 64 0 . 5 1 1 . 5 2 2 . 5 3 P C P C ICA CA R D P R D P GoT o M yP C X X s s s sh

• C

N X N X VN C S u n S u n R ay TH IN C

p e p er w eb eb page page l a tency ency (s )

802.11g PDA web performance 802.11g PDA web performance

0 . 84 0 . 48 3 . 15 0 . 46 0 . 46 0 . 5 1 1 . 5 2 2 . 5 3 P C P C ICA CA R D P R D P GoT o M yP C X X s s s sh

• C

N X N X VN C S u n S u n R ay TH IN C

p e p er w eb eb page page l a tency ency (s )

desktop LAN A/V performance desktop LAN A/V performance

100 % 28 % 15 % 100 % 76 % 12 % 14 % 13 % 100 % N/A 0 % 0 % 2 0 2 0 % 4 0 4 0 % 6 0 6 0 % 8 0 8 0 % 1 0 0 1 0 0 % P C P C ICA CA R D P R D P GoT o M yP C X X s s s sh

• C

N X N X VN C S u n S u n R ay TH IN C

aud aud i

• /v
• /v i

deo deo qua qua l i l i ty

desktop WAN A/V performance desktop WAN A/V performance

100 % 20 % 2% 1% 65 % 27 % 12 % 8% 12 % 100 % 0 % 0 % 2 0 2 0 % 4 0 4 0 % 6 0 6 0 % 8 0 8 0 % 1 0 0 1 0 0 % P C P C ICA CA R D P R D P GoT o M yP C X X s s s sh

• C

N X N X VN C S u n S u n R ay TH IN C

aud aud i

• /v
• /v i

deo deo qua qua l i l i ty

802.11g PDA A/V performance 802.11g PDA A/V performance

7% 19 % 1% 14 % 100 % 0 % 0 % 2 0 2 0 % 4 0 4 0 % 6 0 6 0 % 8 0 8 0 % 1 0 0 1 0 0 % P C P C ICA CA R D P R D P GoT o M yP C X X s s s sh

• C

N X N X VN C S u n S u n R ay TH IN C

aud aud i

• /v
• /v i

deo deo qua qua l i l i ty

thinc thinc

 ultra-thin client

leverage and virtualize standard display driver interface fast, lightweight full-motion, full resolution audio/video performance

transparent transparent checkpoint/migration checkpoint/migration

checkpoint checkpoint

 capture the state of a running

process and save it so that it can be resumed at a later time

migration migration

 move checkpointed process state to a

target machine and resume process

transparent transparent

 no application changes  no need for new languages/run-time  no operating system kernel changes  no constraints on use of OS services

enables enables

 software mobility  load balancing  power management  fault resilience  improved system availability

approaches approaches

language Emerald library Condor kernel Mosix hardware Vmware, Xen system call zap

Hardware Libraries Applications Operating System

zap zap

 virtualize OS to decouple

applications from underlying OS instance

 use high-level kernel functionality

for portable migration

 preserve application availability

across operating system upgrades

problem problem

int iChildPID; if (iChildPID=fork()) { /* parent does some work */ waitpid(iChildPID); } else { /* child does some work */ exit(0); }

issues issues

 resource consistency  resource conflicts  resource dependencies  transparency

solution solution

private virtual namespace

 PrOcess Domain (POD)

POD virtualization POD virtualization

 PID  IPC  file system  network  devices

architecture architecture

 POD virtualization +  checkpoint/restart  kernel module  in Linux

cost? cost?

virtualization virtualization

0.5 1 1.5 2 2.5 3 3.5 4 getpid fork hackbench httpperf make mysql volano

Normalized latency

Xen Zap Linux SMP Zap SMP

remote desktop remote desktop

checkpoint/restart checkpoint/restart

19 0 . 851 0 . 942

2 4 6 8 1 0 1 0 1 2 1 2 1 4 1 4 1 6 1 6 1 8 1 8 2 0 2 0 no nor m a l s ta r t r tup up zap zap checkpo checkpo i n t zap zap re s e sta r t r t t i m e ( s )

zap zap

 transparent checkpoint/migration

• f legacy and network applications

 POD: consistent, conflict-free,

avoid dependencies fast and lightweight

conclusions conclusions

technology scaling trends are driving thin-client computing key enabling technologies: remote display and checkpoint/migration THINC and Zap: display and

• perating system virtualization

mechanisms for thin clients

the future the future

 virtual computing utility

delivered to smart displays

more info more info

network computing laboratory

http://www.ncl.cs.columbia.edu