Application protocols & presentation layer

Transcription

1 Application protocols & presentation layer Suguru Yamaguchi NAIST Overview = Application model = Presentation layer = Large scale services 1

2 TCP/IP Protocol Suites OSI TCP/IP Processing unit Identification Application Presentation Application Message / Stream Session Transport Transport Transport Packet Port Network Internet IP Datagram IP address Data Link Network Interface Frame Datalink Address Physical Hardware Upper layer protocols = Protocols over Transport Layer are called Upper Layer Protocols. Defined by applications Designed mostly by technical requirements by the application. 2

3 Session Layer = Definition for semantic data unit for data-processing Transaction message Session = Definition of data- processing Transaction Logging & Roll-back operation Session Termination = Definition of communication from the model of application Presentation Layer = Data expression in the communication Common data expression across any platform, without ambiguities. Ex. Numeric value 1 to be transmitted, but how? Byte length?» 1, 2, 4, less than 1 byte (6 bits),. Byte order?» Little Endian / Big Endian Bit order?» MSB first, LSB first 3

4 Application Layer = Protocols for applications, but not application itself. SMTP (simple mail transfer protocol) for s applications uses SMTP MTA: sendmail, qmail, postfix, etc. MUA: Eudora, Mozilla Thunderbird, MS/Outlook, etc. Application model 4

5 Viewpoints of application model = Communication model Client / Server model Broadcast based parallelization Peer-to-Peer (P2P) model = Processing model Side effect of communications Process semantics Atomic / checkpoint = Functions End user services Network management Client/Server client Request/reply server client 5

6 Why client / server first? = Easy to understand User interfaces by a client Functions provided by the shared server Easy for development and operation = Easy for service discovery Small number of servers Well-known address (server + well-known port Very hard for dynamic binding Semi-dynamic scheme = DNS Still DNS is popular, even dynamic binding is available Parallelization using broadcast = Broadcasting periodically for data sharing 6

7 Why broadcast? = Send a data in one transmission only for up & running nodes, using UDP Limitation of broadcast All in the same L2 segment. = Not only for any host in multiple L2 segment needs TCP Server federation server client Request/reply server server 7

8 Peer-to-Peer (P2P) / overlay network = Self organized communication configuration = Automatic service discovery Side effect = Communication failure Packet loss Packet duplication Retransmission = Side effect Do something by the request sent from clients Idempotent semantics Exactly once semantics At least once semantics accept errors) At most once semantics accept retransmissions) 8

9 Parallel processing by cluster = Easy for parallelization for idempotent processing = Separate no side effect part and with side effect parts. Virtual single server The Internet Idempotent component With side effects DNS 9

10 How do I find the other end? Three well-known patterns of finding the peer: = 1. the other end is known a priori. = 2. central authority knows the other end. = 3. the other end has to be somebody who processes this information. 1. A priori knowledge of endpoint = A name or IP address is provided in advance = e.g., via Uniform Resource Locator (RFC1738) Type of application protocol TCP port number Host name etc. \_/ \ /\ / \ / \ / scheme authority path query fragment 10

11 Addressing vs naming: its resolution naist.jp Named objects Name resolution Addressable objects 2001:200:16a:8::230 DNS: domain name system = Single global tree for hostname to IP resolution = Hierarchical delegation Root. gtld: generic top-level domain.com.net.org.biz.info cctld: country code top-level domain.jp.th.ch.fr Secondary level domains co.jp, ac.jp, go.jp 11

12 DNS: distributed DB of records. IN NS a.root-servers.net Root zone jp. IN NS a.dns.jp.jp zone naist.jp. IN NS ns.naist.jp naist.jp zone www IN A IN AAAA 2001:200:16a:8::230.com zone Hands on: observe DNS in action In the provided virtual machine: = Run wireshark = Visit some websites = Observe DNS protocol interactions = Some DNS jargons: NS: name server A: address AAAA: IPv6 address 12

13 2. A central authority for endpoint lookup search string tables Endpoint info = Many algorithms for scaling beyond 1000 servers = For further study: NSDI, SIGCOMM papers 3. Discovery of endpoints from information types = Discovery of services associated with the resource Resource identifier Resolver Service endpoint identifier, Capabilities Modern example: OpenID? 13

14 Presentation layer Tech, so far. = Non structured Binary ASCII = Structured TLV: Type, Length, Value ASN.1 XML 14

15 Trade-off capability for data representations XML ASN.1 ASCII TLV binary Bandwidth use efficiency Binary & ASCII = Binary IP header (RFC791), TCP header (RFC793) clear definition on host bit order and network bit order. proper numeric integer handling = ASCII FTP (RFC959) SMTP (RFC2821) POP3 (RFC1939) also for not 8bit clean channel 15

16 FTP : 220 ftp.isi.edu NcFTPd Server (free educational license) ready : USER anonymous : 331 Guest login ok, send your complete address as password : PASS : 230 Logged in anonymously : SYST : 215 UNIX Type: L : PWD : 257 "/" is cwd : TYPE I : 200 Type okay : CWD /in-notes : 250 "/in-notes" is new cwd. SMTP 1758: 220 ns.ixj-mc.com ESMTP Sendmail 8.9.3p2+3.1W/3.7W/ns; Fri, 30 May :15: : EHLO mf.aist-nara.ac.jp 1758: 250 ns.ixj-mc.com Hello 168.pool3.ftthtokyo.att.ne.jp [ ], pleased to meet you 0025: MAIL FROM:<youkix@is.naist.jp> SIZE= : 250 <youkix@is.naist.jp>... Sender ok 0025: RCPT TO:<bunji@iij-mc.com> 1758: 250 <bunchan@ixj-mc.com>... Recipient ok 0025: DATA 1758: 354 Enter mail, end with "." on a line by itself 0025: Received: from mf.aist-nara.ac.jp (localhost [ ]) (JST) 16

17 TLV = Type, Length, Value = OSPF (RFC2328) = RADIUS (RFC2138) = Structured expression of data Type invokes a proper process. Length is for variable data. Value is depending to protocols. Function rich mechanism comes way = XDR (extended Data Representation) for Sun RPC/ NFS (1980 s) = ASN.1 (1980 s) = XML (1990 s) = Requirements (desired characteristics) good wire efficiency easy to handle fluent for data expression, on static typing and name space. powerful tools and rich libraries 17

18 ASN.1 (BER encoding) = (tag, length, value) Tag: ASN.1 type Length: size of the ASN.1 value Value: ASN.1 value = type of ASN.1 INTEGER OCTET STRING OBJECT IDENTIFIER SEQUENCE (array) SNMP Object Identifier = Iso(1).org(3).dod(6).internet(1) Mgmt (2) Experimental (3) Private (4) = data structure definition for SNMP MIB-II (RFC1214), Structure of Management Information version 2 (SMIv2) (RFC2578) 18

19 XML = <?xml version = "1.0"?> = <tag attribute="value"> = <another-tag another-attribute="value" /> = </tag attribute="value"> XML namespace <?xml version='1.0' encoding='utf-8'?> <soap:envelope xmlns:soap=' xmlns:xsi=' xmlns:xsd=' xmlns:soapenc=' soap:encodingstyle=' <soap:body> <n:getquoteresponse xmlns:n='urn:xmethods-delayed-quotes'> <Result xsi:type='xsd:float'>7.92</result> </n:getquoteresponse> </soap:body> </soap:envelope> 19

20 XML data type XML Schema <?xml version="1.0" encoding="utf-8"?> <SOAP-ENV:Envelope xmlns:xsi=" xmlns:soap-enc=" xmlns:xsd=" SOAP-ENV:encodingStyle=" xmlns:soap-env=" <SOAP-ENV:Body> <namesp1:getquote xmlns:namesp1="urn:xmethods-delayed-quotes"> <symbol xsi:type="xsd:string">akam</symbol> </namesp1:getquote> </SOAP-ENV:Body> </SOAP-ENV:Envelope> XML more goodies = data description: RDF = tools: XPath, XSLT, = API: DOM, SAX, etc. = digital signature & encryption XML Digital Signature XML Encryption 20

21 XML is very important = XML becoming very popular in many information systems. data set in XML programming for XML data process = service systems with network capability use XML as its standard way for data manipulations. not only for service system programming XML LISA: Large Information System Architecture 21

22 Why we need LISA? = many clients, few servers = performance bottleneck Virtual single server The Internet Idempotent component With side effects Load balancing = a mechanism for load balancing = LB; Load Balancer) = Ideas behind. LB manages termination points of TCP connections. Client: Server1: The Internet vurtual server Server2: Forwarding Table Src: Dst: [1-3] Server3:

23 Termination point management = Which server takes the connection? round robin, least connection, performance oriented, failure mana how to know individual server down asap? = How LB handles complete server address dependency? simple NAT is too simple, not working well Cookies and HTTP payloads contain some IP address inside. not idempotent for all More innovations = Any server does not have to be in the same L3 segment. combine multiple servers scattering over the Internet and merge them virtually as a single server, using tunneling etc. Sever federation = CDN (Contents Delivery Network) load balancing among ISPs CDN Akamai.com is s pionieer. 23

24 Providing Service at your site. server IX clients Contents Delivery Network latest contents feed from DC (freshness control) Data Center clients feed for customers from CD server inside the ISP (well-managed access) 24

25 Anycast = load balancing using routing management = Root DNS servers Anycast HOS T 2001::1 HOST HOST HOST 2001::1 2001::1 25

26 Geographical distribution using Anycast = F root DNS Originally at ISC in Silicon Valley, Calif. Currently, Hong Kong & Amsterdam??? Requests from clients travel to the nearest host. Hong Kong, CN same routing information injected by servers. Amsterdam, NL Silicon Valley, Calif. 26