Transcript Unit 10

Operating Systems
Operating Systems
Unit 10:
– Networking
– Distributed File Systems
Networking concepts
• host
– single computer
– local CPU, memory, disks
• link
– connecting media
– attributes:
• bandwidth, latency
COP 5994 - Operating Systems
2
Network Topologies
COP 5994 - Operating Systems
3
Communication Protocols
Application
Data Transport
COP 5994 - Operating Systems
4
Data Transport layers
4: Transport layer
• End-to-end communication
• Relies on network layer to determine proper path
from one end of communication to the other
3: Network layer
• Moving data between computers
2: Link layer
• Provides an interface between the network layer
and the underlying physical medium of the
connection
1: Physical layer
COP 5994 - Operating Systems
5
2: Link Layer
• first software layer on top of physical
medium
• unit: frame
• responsibilities:
– deliver frame
– detect errors
– correct transmission errors
COP 5994 - Operating Systems
6
2: IEEE 802.3 CSMA/CD
(Ethernet)
• Carrier Sense Multiple Access
with Collision Detection protocol
– transceiver tests delivery medium for availability
• due to delays in medium, it is possible that multiple
transceivers begin transmitting simultaneously
– if transceivers detect collision:
• continues to transmit bytes for a specific period of
time to ensure that all transceivers become aware of
the collision
• waits for a random interval before attempting to
retransmit
COP 5994 - Operating Systems
7
2: Token Ring
• operates on ring networks
• employs tokens to gain access to the
transmission medium
• token is empty frame that is circulated
between machines
COP 5994 - Operating Systems
8
2: Token Ring example scenario
1/2
COP 5994 - Operating Systems
9
2: Token Ring example scenario
2/2
COP 5994 - Operating Systems
10
2: Fiber Distributed Data Interface
(FDDI)
• operates over fiber-optic cable
– support more transfers at greater speeds
over larger distance
• built on two token rings
– the second usually being reserved for
backup
COP 5994 - Operating Systems
11
2: IEEE 802.11 (Wireless)
• similar to Ethernet:
– Carrier Sense Multiple Access
with Collision Avoidance (CSMA/CA)
– require each sender broadcast a Request to
Send (RTS) to entire network
– upon receiving an RTS
• receiver broadcasts a Clear to Send (CTS)
message to the entire network if the medium is
available
COP 5994 - Operating Systems
12
3: Network Layer
•
•
routes packet to next host toward destination
determines next host
– address information
– network topology
– link quality
• strength of signal, error rate and interference
– Interference is broadcast throughout networks using
various router protocols, such as Routing Information
Protocol (RIP)
COP 5994 - Operating Systems
13
3: Internet Protocol (IP or
IPv4)
• hosts have unique IP address
– 32-bit number written as quad
– network
• first n bits of IP number, written as “/n”
• 8 - class A, 16 - class B, 24 - class C
• more than 24 - class D
– netmask
• 32 bit number with first n bits all 1, rest 0
– broadcast
• network number (first n bits), rest all 1
– gateway IP
COP 5994 - Operating Systems
131.94.134.39
/24
255.255.255.0
131.94.134.255
131.94.134.1
14
3: Internet Protocol (IP or
IPv4)
• host names are mapped to IP address
via the Domain Name System (DNS)
– need IP numbers of DNS servers
• private IP numbers
– used locally
– Network Address Translation (NAT)
COP 5994 - Operating Systems
15
3: IPv4 packet
COP 5994 - Operating Systems
16
3: Internet Protocol version 6
(IPv6)
• larger address space: 128 bit number
– written as 8 groups of 4 hex digits
2001:468:701:3800:206:5bff:fe53:2a87
COP 5994 - Operating Systems
17
3: IPv6 new features
• quality of service
• authentication and privacy
• anycast address
– used to send packet to one of a group
• multicast
– used to send packet in all of a group
COP 5994 - Operating Systems
18
4: Transport Layer
• Connection-oriented approach
• Hosts send each other control information
through handshaking to set up a logical end-toend connection
• Imposes reliability on unreliable networks
• Guarantees that data sent from sender will arrive
at intended receiver undamaged and in correct
sequence
• Connectionless approach
• Two hosts do not handshake before transmission
• No guarantee that sent messages will be
received in their original order, or at all
COP 5994 - Operating Systems
19
4: Transmission Control Protocol
(TCP)
• connection-oriented transmission
protocol
– guarantees that segments sent from a
sender will arrive at the intended receiver
undamaged and in correct sequence
– handles error control, congestion control,
and retransmission
– allows protocols like HTTP and FTP to send
information into network as simply and
reliably as writing to a file on the local
computer
COP 5994 - Operating Systems
20
4: User Datagram Protocol (UDP)
• Connectionless User Datagram Protocol
– provides minimum overhead necessary for
the transport layer
– no guarantee that UDP datagrams will reach
their destination in their original order, or at
all
COP 5994 - Operating Systems
21
Application Layers
• Application layer protocols
– Specify rules that govern remote
interprocess communication
– Determine how processes should interact
• protocols interact with remote resources
– via Uniform Resource Identifier (URI)
COP 5994 - Operating Systems
22
Application Layer protocols
URL: Uniform Resource Locator
– special form of URI
syntax: protocol://host:port/path
• Common protocols
– ftp
– http
– smtp
– ldap
– sip
COP 5994 - Operating Systems
23
Application Layer protocol: RPC
Remote Procedure Call
general mechanism to enable function
invocation across host boundaries
• extension of local function call model
COP 5994 - Operating Systems
24
Application Layer protocol: RPC
COP 5994 - Operating Systems
25
RPC portmapper
• common Unix implementation
• translates RPC program numbers
– TCP/IP port numbers
– UDP/IP port numbers
• common RPC programs
– NIS
– NFS
COP 5994 - Operating Systems
26
Distributed File Systems
• Networked file systems
– Allow clients to access files stored on
remote computers
• Distributed file systems
– Special examples of networked file systems
that allow transparent access to remote files
COP 5994 - Operating Systems
27
Distributed File System Concepts
• location transparency
– user is unaware of the physical location of a file
within a distributed file system
– user sees only a global file system
• caching & consistency
– Clients keep a local copy of a file and flush modified
copies of it to the server from time to time
– Because there are multiple copies of the same file,
files can become inconsistent
COP 5994 - Operating Systems
28
Distributed File System Concepts
• scalability
– Distributed file systems are designed to share
information among large groups of computers
– New computers should be able to be added to the
distributed system easily
• security
– Ensuring secure communications
– Access control
• fault tolerance
COP 5994 - Operating Systems
29
Distributed File System Concepts
• server state
– can be either stateful or stateless
– stateful system
• server keeps state information of the client
requests so that subsequent access to the file is
easier
– stateless system
• client must specify which file to access in each
request
COP 5994 - Operating Systems
30
Network File System
• developed by Sun MicroSystems
• early 1980s
• dominant for Unix environments
• Versions
– NFS-3
– NFS-4, IETF standard
COP 5994 - Operating Systems
31
NFS Architecture
COP 5994 - Operating Systems
32
Network File System (NFS)
• NFS versions 2 and version 3
– assume a stateless server implementation
– if the server crashes:
• client simply retries until the server responds, or
• aborts
– if server resumes, no need to rebuild state
COP 5994 - Operating Systems
33
Network File System (NFS)
• NFS-4
– Stateful
– enables faster access to files
– if the server crashes
• all the state information of the client is lost
• client needs to rebuild its state on the server
before retrying
COP 5994 - Operating Systems
34
NFS-4 Caching: Delegation
• efficient client-caching scheme:
– server temporarily transfers control of file to
client
• read delegation: no other client can write
• write delegation: no other client can read or write
– If another client requests a delegated file,
the server will revoke the delegation and
request that the original client flush the file
back to server
COP 5994 - Operating Systems
35
Andrew File System
• developed by Carnegie Mellon University
– after first names of A. Carnegie & A. Mellon
– early 1980s
• design goals:
– large scale
– secure
– efficient
• Versions: AFS-1, AFS-2, AFS-3
– basis for DFS of OSF, predecessor to Coda
COP 5994 - Operating Systems
36
Andrew File System (AFS)
• location transparency
– all files appear as a branch of a traditional
UNIX file system at each workstation
• built on RPC
• security:
– uses Kerberos authentication server
– access control lists for file/directories
COP 5994 - Operating Systems
37
AFS Structure
COP 5994 - Operating Systems
38
AFS caching
• files are cached on clients local disk
– notifies clients via callback that files are no
longer valid
– client must invalidate its file and request the
most recent version
COP 5994 - Operating Systems
39
Coda File System
• developed by Andrew team
• design goal: fault tolerance
• mid 1980s
COP 5994 - Operating Systems
40
Coda File System Concepts
• Volumes
– logical pieces of the file system
– replicated physically across multiple file servers
• Volume storage group (VSG)
– servers that hold the same volume
• Available volume storage group (AVSG)
– reachable members of the VSG
• Preferred Server
– member of AVSG, currently serving files
COP 5994 - Operating Systems
41
Coda volume structure
COP 5994 - Operating Systems
42
Coda read/write scenario
• to read, check cache for file
– if found, done
– else
• get file from PS
• get file versions from other members of AVSG
• if conflict, then AVSG members agree on new
versions, and update their copies
• to write,
– send file to all members of AVSG
– track members of AVSG that have written
file
COP 5994 - Operating Systems
43
Coda inconsistency scenario
COP 5994 - Operating Systems
44
Coda client disconnect
scenario
• When connected to Coda, clients cache files
so they can be accessed when disconnected
(hoarding stage)
• When disconnected, clients enter the
emulation stage where all file requests are
serviced from the cache, if the file is resident
(error otherwise)
• When reconnected, file updates are sent to
the server asynchronously (reintegration
stage)
COP 5994 - Operating Systems
45
Sprite File System
• Part of Sprite distributed OS
– Developed at UC Berkeley, mid 1990s
• Sprite file system characteristics
– Emulates a UNIX file system
• even remote access to I/O devices
– Every client has the exact same view of the
hierarchy
COP 5994 - Operating Systems
46
Sprite File System domains
per client
COP 5994 - Operating Systems
47
Sprite read scenario
– to open a file, the client first checks its cache,
then makes a request to server
– if the server is unable to satisfy the request from
its cache, it reads the data from disk
– both caches retain a copy of file
COP 5994 - Operating Systems
48
Sprite write scenario
– client writes to its cache
– updated pages are flushed to server every 30
seconds
– lazy write-back strategy
COP 5994 - Operating Systems
49
Sprite caching protocol
• sequential write sharing
– upon opening file for write, client checks file
version with server
• might cause reload of cached file from server
• concurrent write sharing
– if 2 clients open file for write, caching is
turned disabled
COP 5994 - Operating Systems
50
Common Internet File System
• started by IBM,
no propagated by Microsoft
• other name: server message block (smb)
• built on top of NetBIOS
– name service
– session management
– datagram distribution
• also supports printing
COP 5994 - Operating Systems
51
CIFS concepts
• Universal Naming Convention (UNC)
\\host\path\file
• server publishes shares
– workgroup elects browse master
• client maps share to drive letters
COP 5994 - Operating Systems
52
CIFS caching
• enabled via locking concept
• opportunistic lock (oplock)
– exclusive oplock
• client is granted abritrary buffering rights
– batch oplock
• client keeps lock despite no local accessor
– level II oplock
• multiple clients read, no writer
COP 5994 - Operating Systems
53
CIFS features
• user/share authentication
– plain password
– challenge/response protocol
with password encryption
• replicated virtual volumes
– volumes may span servers
– volumes can be transparently moved
– volumes may be replicated
COP 5994 - Operating Systems
54
Agenda for next week:
– Chapter 19:
Security
– Read ahead !
COP 5994 - Operating Systems
55