Content Distribution Networks

Transcription

1 ontent Distribution Networks Outline Implementation Techniques Hashing Schemes edirection Strategies Spring 22 S Design Space aching explicit transparent (hijacking connections) eplication server farms geographically dispersed (DN) Spring 22 S Story for DNs Traditional: Performance move content closer to the s avoid server bottlenecks New: DDoS Protection dissipate attack over massive resources multiplicatively raise level of resources needed to attack Spring 22 S

2 Denial of Service Attacks (DoS) attacker server Spring 22 S Distributed DoS (DDoS) zombie attacker server slave attacker zombie zombie zombie Spring 22 S edirection Overlay Geographically distributed server clusters Internet Backbone s Distributed request-redirectors Spring 22 S

3 Techniques DNS one name maps onto many addresses works for both servers and reverse proxies HTTP requires an extra round trip outer one address, select a server (reverse proxy) content-based routing (near ) UL ewriting embedded links Spring 22 S edirection: Which eplica? Balance Load ache Locality Network Delay Spring 22 S Hashing Schemes: Modulo UL (key) % svr svrn Easy to compute Evenly distributed Good for fixed number of servers Many mapping changes after a single server change Spring 22 S

4 onsistent Hashing (Hash) svrn Unit circle url- svr svr1 url-1 svr2 Hash server, then UL losest match Only local mapping changes after adding or removing servers Used by State-of-the-art DNs Spring 22 S UL svr svr1 svr2 svr svr svrn Highest andom Weight (HW) H sort weight weight1 weight2 weight weightn high Hash(url, svraddr) Deterministic order of access set of servers Different order for different ULs Load evenly distributed after server changes low Spring 22 S edirection Strategies andom (and) equests randomly sent to cooperating servers Baseline case, no pathological behavior eplicated onsistent Hashing (-Hash) Each UL hashed to a fixed # of server replicas For each request, randomly select one replica eplicated Highest andom Weight (-HW) Similar to -Hash, but use HW hashing Less likely two ULs have same set of replicas Spring 22 S

5 edirection Strategies (cont) oarse Dynamic eplication (D) Using HW hashing to generate ordered server list Walk through server list to find a lightly loaded one # of replicas for each UL dynamically adjusted oarse grained server load information Fine Dynamic eplication (FD) Bookkeeping min # of replicas of UL (popularity) Let more popular UL use more replicas Keep less popular UL from extra replication Spring 22 S Simulation Identifying bottlenecks Server overload, network congestion End-to-end network simulator prototype Models network, application, and OS Built on NS + LAD simulators 1s of servers, 1s of s >6, req/s using full-tp transport Measure capacity, latency, and scalability Spring 22 S WA S PA A O SD A S Network Topology S S S MI MA IL NE D S S S GA TX S S Server, lient, - outer Spring 22 S

6 Simulation Setup Workload Static documents from Web Server trace, available at each cooperative server Attackers from random places, repeat requesting a subset of random files Simulation process Gradually increase offered request load End when servers very heavily overloaded Spring 22 S apacity: 64 server case Normal Operation and -Hash -HW D FD FD-Ideal A single server can handle ~6 req/s in simulation Spring 22 S apacity: 64 server case Under Attack (25 zombies, 1 files, avg 6KB) and -Hash -HW D FD FD-Ideal A single server can handle ~6 req/s in simulation Spring 22 S

7 DF of esponse Latency: 64 Servers Under Attack andom s Max: 11.2k req/s -HW -Hash FD D and esponse Time in Logscale (Seconds) DF of esponse -Hash Max: 19.8k req/s Spring 22 S HW FD -Hash D esponse Time in Logscale (Seconds) Latency At D s Max: 35.1k req/s FD D DF of esponse esponse Time in Logscale (Seconds) Spring 22 S apacity Scalability Normal Operation Num of Servers D -HW and Under Attack (25 zombies, 1 files) Num of Servers D -HW and Spring 22 S

8 Various Attacks (32 servers) 1 victim file, 1 KB 1 victim files, avg 6KB Num of Zombies (slave attackers) and -HW D Num of Zombies (slave attackers) and -HW D Spring 22 S Deployment Issues Servers join DDoS protection overlay Same story as Akamai Get protection and performance lients use DDoS protection service Same story as proxy caching Incrementally deployable Get faster response and help others Spring 22 S