INF 5071 Performance in Distributed Systems Introduction Motivation

INF 5071 – Performance in Distributed Systems Introduction & Motivation 1/9 - 2006

Overview ü About the course ü Application and data evolution ü Architectures ü Machine Internals ü Network approaches ü Case studies INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

INF 5071: The Course

Lecturers ü Carsten Griwodz Ø email: griff @ ifi Ø office: Simula 153 ü Pål Halvorsen Ø email: paalh @ ifi Ø office: Simula 132 INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Content architectures Networ k file systems Networ k distribution Networ k resource scheduling Networ k protocols topologies INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Content ü Applications and characteristics (components, requirements, …) ü Server examples and resource management (CPU and memory management) ü Protocols with and without Quality of Service (Qo. S) (specific and generic Qo. S approaches) INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Content ü Storage systems (management of files, retrieval, …) ü Distribution (use of caches and proxy servers) ü Peer-to-Peer (various clients, different amount of resources) ü Guest lecture: The FAST searching system (architecture: resource utilization and performance, storage and distribution of data, parallelism, etc. ) INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Content - student assignment ü Mandatory student assignment (will be presented more in-depth later): Ø write a project plan describing your assignment write a report describing the results and give a presentation Ø for example (examples from earlier): Ø n n n n n STI Cell processor toolkit setup TCP friendliness variations DCCP setup and tests Network emulators Comparison of Linux schedulers (cpu, network, disk) File system benchmarking (different OSes and file systems) Comparison of methods for network performance monitoring (packet train, packet pair, ping, tcp dump library/pcap, …) Compare media players (VLC, mplayer, xine, …) Compare multimedia pipelines (gstreamer, nmm, …) … INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Goals ü Distribution system mechanisms enhancing performance Ø Ø Ø architectures system support protocols distribution mechanisms … ü Be able to evaluate any combination of these mechanisms INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Exam ü Prerequisite: approved presentation of student assignment ü Oral exam (early December 2006): Ø all transparencies from lectures Ø content of your own student assignment INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Evolution

Discrete Data to Continuous Media Data INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Evolution of (continuous) media streams: Television (Broadcast) channels time sender • analog or digital • traditionally, one program per channel receiver(s) q analog use frequency division multiplexing only q digital may additionally use time division multiplexing inside one frequency (several programs per channel) INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Evolution of (continuous) media streams: Near Video-on-Demand (NVo. D) channels time sender • analog or digital broadcasting • one program over multiple channels • time-slotted emission of the program INF 5071 – performance in distributed systems receiver(s) 2006 Carsten Griwodz & Pål Halvorsen

Evolution of (continuous) media streams: (True) Video-on-Demand (Vo. D) movies time sender receiver(s) • digital uni- or multicasting • control channels INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Evolution of (continuous) media streams: “Interactive Vision” data stream time sender receiver(s) • digital uni- or multicasting • control channels • fixed non-linear data streams INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Evolution of (continuous) media streams: “Cyber Vision” time sender • digital uni- or multicasting • control channels • variable non-linear “media”, e. g. , - games, virtual reality, … INF 5071 – performance in distributed systems receiver(s) 2006 Carsten Griwodz & Pål Halvorsen

Evolution & Requirements: File download and Web browsing Internet Packet loss Not acceptable Bandwidth demand Low (? ) Accepted delay Medium – High (? ) Synchronicity - INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Evolution & Requirements: Live and on-Demand Streaming Packet loss Acceptable Bandwidth demand High Accepted delay Medium Synchronicity Internet INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Evolution & Requirements: Textual commands and textual chat Internet Packet loss Not acceptable Bandwidth demand Low Accepted delay Human reading speed Synchronicity - INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Evolution & Requirements: AV chat and AV conferencing Packet loss Acceptable Bandwidth demand High Accepted delay Medium Synchronicity Lip synchronous Internet INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Evolution & Requirements: Haptic Interaction Internet Packet loss Acceptable Bandwidth demand Low Accepted delay Human reaction time Synchronicity - INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Evolution & Requirements: A distributed system must support all Internet INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Different Views on Requirements ü Application Ø Qo. S – time sensitivity? Ø resource capabilities – bandwidth, latency, loss, reliability, … Ø best possible perception ü Business Ø scalability Ø reliability Æ Architectural Ø topology Ø cost vs. performance INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Components ü Servers ü End-systems Ø PCs Ø TV sets with set-top boxes Ø PDAs Ø Phones Ø … ü Intermediate nodes Ø routers Ø proxy cache servers ü Networks Ø backbone Ø local networks INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Technical Challenges ü Servers (and proxy caches) Ø storage n continuous media streams, e. g. : o o n n Ø * 90 minutes * 15 Mbps (HDTV) * 74 minutes * 1. 4 Mbps = 40. 5 TB = 1. 4 TB metrological data, physics data, … web data – people put everything out nowadays I/O n n n many concurrent clients real-time retrieval continuous playout o o n DVD (~4 Mbps) HDTV (~15 Mbps) current examples of capabilities o o o Ø 4000 movies 2000 CDs disk: Seagate X 15 - ~400 Mbps network: Gb Ethernet (1 and 10 Gbps) bus(ses): - PCI 64 -bit, 133 Mhz (8 Gbps) - PCI-Express (2. 5 Gbps each direction/lane) computing in real-time n n encryption adaptation transcoding … INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Technical Challenges ü User end system Ø real-time processing of data (e. g. , 1000 MIPS for an MPEG-II decoder) Ø storage of media/web files Ø request/response delay (< 150 ms for videophones) Ø high data rates, e. g. , MPEG-II DVD quality: n n n Ø max. total video data rate of ~10 Mbps average transport stream of 4 – 8 Mbps (video, audio, headers, error protection) max. user rate of ~11 Mbps (all included like control signals) more challenging if client contributes and share its resources with the rest of the system in a P 2 P manner ü Network Ø real-time transport of media data Ø high rate downloads Ø TCP fairness Ø mobility Ø … INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Traditional Architectures

Client-Server ü Traditional distributed computing ü Successful architecture, and will continue to be so (adding proxy servers) ü Tremendous engineering necessary to make server farms scalable and robust backbone network local distribution network INF 5071 – performance in distributed systems local distribution network 2006 Carsten Griwodz & Pål Halvorsen

Server Hierarchy ü Intermediate nodes or proxy servers may offload the main master server completeness of available content master servers ü Popularity of data: not all are equally popular – most request directed to only a few (Zipf distribution) regional servers ü Straight forward hierarchy: Ø Ø popular data replicated and kept close to clients locality vs. communication vs. server costs local servers end-systems INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Peer-to-Peer (P 2 P) ü Really an old idea - a distributed system architecture Ø Ø Ø ü No centralized control Nodes are symmetric in function All participating and sharing resources Typically, many nodes, but unreliable and heterogeneous backbone network local distribution network INF 5071 – performance in distributed systems local distribution network 2006 Carsten Griwodz & Pål Halvorsen

Topologies ü Client / server Ø easy to build and maintain Ø severe scalability problems ü Hierarchical Ø complex Ø potential good performance and scalability Ø consistency challenge Ø cost vs. performance tradeoff ü P 2 P Ø complex Ø low-cost (for content provider!!) Ø heterogeneous and unreliable nodes ü We will in later lectures look at different issues for all these INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Traditional Server Machine Internals

General Operating System Structure and Retrieval Data Path application user space kernel space file system INF 5071 – performance in distributed systems communication system 2006 Carsten Griwodz & Pål Halvorsen

Example: Intel Hub Architecture (850 Chipset) – I Intel D 850 MD Motherboard: RDRAM connectors CPU socket system bus RDRAM interface hub interface PCI bus Memory Controller Hub I/O Controller Hub PCI connectors INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Example: Intel Hub Architecture (850 Chipset) – II Note: these transfers only show dataapplication movement between sub-systems and not the commands themselves. communication Additionally, data file touching system operations within a subsystem will require that data is moved from memory and to the CPU, e. g. : disk - checksum calculation - encryption network card - data encoding - forward error correction Pentium 4 Processor registers cache(s) system bus (64 -bit, 400/533 MHz ~24 -32 Gbps) memory controller hub RAM interface (two 64 -bit, 200 MHz ~24 Gbps) RDRAM file system RDRAM communication system RDRAM application RDRAM hub interface (four 8 -bit, 66 MHz 2 Gbps) I/O controller hub PCI slots PCI bus (32 -bit, 33 MHz 1 Gbps) INF 5071 – performance in distributed systems network card PCI slots disk 2006 Carsten Griwodz & Pål Halvorsen

Example: IBM POWER 4 application POWER 4 chip CPU L 1 Note: Again, data touching file systemoperations add movement operations CPU L 1 disk core interface switch ) ps L 2 0 M 0 , 4 bit 4 - 6 ur (fo L 3 controller ~9 H 0 M 0 , 4 bit z u (fo L 3 ~9 (e 3 ht ig memory controller Hz ~9 0 M 0 t, 4 bi 2 - - 4 r 6 ) ps b 5 G RAM file system communication system RAM application RAM PCI busses GX bus (two 32 -bit, 600 MHz ~35 Gbps) remote I/O (RIO) bridge Hz network card ) ps b 5 G fabric controller GX controller communication system (32/64 -bit, 33/66 MHz 1 -4 Gbps) PCI host bridge PCI-PCI bridge PCI slots network card PCI slots disk RIO bus (two 8 -bit, 500 MHz ~7 Gbps) INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Example: AMD Opteron & Intel Xeon MP 4 P servers application file system communication system disk network card F Know your hardware – different configuration may have different bottlenecks INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Server Internals ü Data retrieval from disk and push to network Ø buffer requirements Ø bus transfers Ø CPU usage Ø Ø concurrent users can be merged? storage (disk) system: n n Ø … scheduling – ensure that data is available in time block placement – contiguous, interleaving, striping ü Stable operations: Ø redundant HW Ø multiple nodes ü Much more, e. g. , caching/prefetching, admission control, … ü We will in later lectures look at several of these INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Network Approaches

Network Architecture Approaches ü WAN backbones Ø SONET Ø ATM ü Local distribution network Ø ADSL (asymmetric digital subscriber line) Ø FTTC (fiber to the curb) Ø FTTH (fiber to the home) Ø HFC (hybrid fiber coax) (=cable modem) Ø E-PON (Ethernet passive optical network) Ø … ü Different capabilities Ø loss rate Ø bandwidth Ø possible asymmetric links ATM / SONET backbone network wireless ADSL telephone cable Ø distance load Ø …. Ø INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Network Challenge ü Goals: Ø network-based distribution of content to consumers Ø bring control to users ü Distribution in LANs is more or less solved: OVERPROVISIONING works Ø Ø established in studio business established in small area (hotel/hospital/plane/…) businesses Network INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Network Challenge ü WANs Ø assumptions: n n Ø programs: n n Ø overprovisioning of resources will NOT work no central control of delivery system need for interoperability – not from a single source need for co-operative distribution systems amount of data: n n n n billions of web-pages billions of downloadable articles estimated 65000 movies made in 1995 data from TV-series, sport clips, news, … games and virtual worlds music … INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Network Challenges ü Different applications and data types have different requirements and behavior ü What kind of services offered is somewhat dependent on the used protocols ü We will in later lectures look at different protocols INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Case Studies: Application Characteristics

i. TVP ü Country-wide IP TV and Vo. D in Poland Ø live & Vo. D Ø hierarchical structure with caching n n regional content centers (receiving data from content providers) a number of proxy caches below (handling requests from users) Ø different quality levels of the video – up to 700 Kbps Ø observations over several months INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

i. TVP: Popularity Distribution ü Popularity of media objects according to Zipf, i. e. , most accesses are for a few number of objects ü The object popularity decreases as time goes ü During a 24 -hour period Ø up to 1500 objects accessed Ø ~1200 accesses for the most popular INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

i. TVP: Access Patterns ü Regular days Ø low in the morning, high in the evening Ø typical 30 requests per minute Ø the most popular items had an average of 300 accesses per day, Ø an average total of 11. 500 accesses per day ü Live transmissions Ø higher request rate Ø an average total of 18. 500 accesses per day Ø 20% accesses to the most popular content ü Event transmissions Ø several hundreds accesses per minute during event transmission Ø an average total of 100. 000+ accesses per day Ø 50% accesses to the most popular content INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

i. TVP: Concurrency and Bandwidth ü The number of concurrent users vary, e. g. , for a single proxy cache Ø Ø event: up to 600 regular: usually less than 20 ü Transfers between nodes are on the order of several Mbps, e. g. , Ø event: n n Ø single proxy: up to 200 Mbps whole system: up to 1. 8 Gbps regular: n n single proxy: around 60 Mbps whole system: up to 400 Mbps INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Funcom’s Anarchy Online ü World-wide massive multiplayer online roleplaying game Ø client-server n n n point-to-point TCP connections virtual world divided into many regions one or more regions are managed by one machine INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Funcom’s Anarchy Online ü For a given region in an one hour trace Griwodz et. al found Ø ~175 players Ø Ø average RTT somewhat above 250 ms OK a worst-case delay of 67 s (!) loss results in a players nightmare less than 4 packets per second small packets: ~120 B thins streams Ø Sharing/competing for both server and network resources INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Verdens Gang (VG): News-on-Demand ü Client-server ü Microsoft Media Server protocol (over UDP, TCP or HTTP) ü From a 2 -year log of client accesses for news videos Johnsen et. al. found Ø Approximated Zipf distributed popularity, but more articles are popular Ø Access pattern dependent on time of day and day of week Ø Large bandwidth requirements, i. e. , several GBs per hour INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Application Characteristics ü Movie-on-Demand live video streaming Ø Access pattern according to Zipf Ø high rates Ø many concurrent users Ø extreme peeks Ø timely, continuous delivery ü Games Ø low rates Ø many concurrent users Ø interactive Ø low latency delivery ü News-on-Demand streaming Ø daily periodic access pattern – close to Zipf Ø similar to other video streaming ü … INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Picture Today! Vo. D WWW network ? ? e nc a m for r e p network Live event network P 2 P INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen

Summary ü Applications and challenges in a distributed system Ø different classes Ø different requirements Ø different architectures Ø different devices Ø different capabilities Ø … Ø and it keeps growing!!!! ü Performance issues are important…!!!! INF 5071 – performance in distributed systems 2006 Carsten Griwodz & Pål Halvorsen