Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
12 Cards in this Set
- Front
- Back
|
Filesystem: Persistent Hierarchic name space Api with CRUD Sharing data with access control Concurrent access Mountable filestores |
File attribute record structure File length Creation timestamp Read timestamp Write timestamp Attribute timestamp Reference count ---------------------------- Owner File type Access control list |
|
|
File service requirements THEF CRCS |
Model file service architecture Directory service(Lookup, AddName, UnName, GetNames) Flat file service(CRWD, GetAttributes, SetAttributes) |
|
|
SunNFS Support for: THEF Limited Support for: CRCS
Architecture: Client: App programs -> VFS -> UNIXFS, Other file Systems, NFS Client Server: App program -> VFS -> NFS Server, NFS Client, UNIXFS |
NFS implementation doesn't need to run at System Kernel level: - Coz can run at application server level Advantages of UNIX implementation: - dont need to recompile, shared cache, can access i-nodes and file blocks, security and encryption key
|
|
|
NFS access control and auth - Stateless server - userId, groupId - Kerberos encryption Mounting - mount(remotehost, remotedirectory, localdirectory) - <IP Address, port, file handle>
|
Automounter - Empty mount - table of mount points - simple form of replication - keeps mount table small
Kerberos - used in mount service - checks UserId, GroupId Problems: - cant have multiple users - all remote stores must be remounted when login |
|
|
NFS optimization - Like UNIX file caching: memory buffer, writes defered to next sync
NFSv3 - write-through: immediate - delayed commit: commit() after close |
Server caching does nothing to reduce RPC traffic between client and server
Timestamp based validity check (T - Tc < t)v(Tmclient = Tmserver) NFS Summary: Access:E, Location:N, Concurrency:L, Replication:L, Failure:L, Mobility: N, Performance: G, Scaling:G |
|
|
Names and identifiers: name -> identifier/address names preferred over identifiers -nane services resolve names
|
Namespace requirements: -management of trust - infinite number of names - structured - simple, meaningful names
DNS lookup -> ARP Lookup | ResourceID -> web server->file |
|
|
Recrusive(Client restriction due to security)/Non recursive navigation Caching - previous name resolutions, validity, other server caches |
DNS - internet, caching, 100ms, tld + subdivisions
algorithm: local cache, superior dns name server(another NS, IP)
DNS resource records: A, NS, CNAME, SOA, MX, TXT, PTR, HINFO
Issues: DNS names change often(bad caching), cant change structure
|
|
|
Directory Service: yellow pages Discover Service: directory service + auto updated on changes + discovers client services
GNS: Cache consistency, structure might change, unique directory identifier #633(world), #599(EC), #642(America) |
X500 - ISO, ITU DUAs and DSA (Server agent, User agent) Tree structure Directory Information Tree(Root->France->University... Directory Information Base (Name, Dept, University, City) |
|
|
Clocktime -Dc UTC - Dt
DTreq = DTresp
Berkley algorithm: Average out time |
Clock sync in wireless networks: Message prep -> Time spent in NIC -> Delivery time to app
Lamports algorithm: Happens before a->b If a->b and a's clock is 60 and b's clock is less than a's, then b's new clock is 61, next frame is 61 + whatever the difference was last time
app -> Middleware adjusts -> network |
|
|
Vector clocks -> vector with: 1. VC[i] is logical clock at Pi 2. VCi[j] = k then Pi knows k events have transpired at Pj. Pi knows local time of Pj.
First, increment VCi <- VCi + 1 set m's timestamp to VCi Pj sets its own vector by VCk[k] <- max(VCk[k] , ts(m)[k]
|
Centralised Mutual exclusion algorithm. If 1 asks for 3, then give. if 2 asks for 3, queue it on 3 (dont reply). When 1 releases 3, give it to 2
Distributed - Lower timestamp wins in a tie
|
|
|
Election algorithm:
1. Bully Algorithm - P sends election message to everyone - if no one responds, P is the winner - if a higher up responds, it takes over. Bye bye P. Higher up becomes co-ordinator
|
Ring algorithm: Collects nodes till target node. If a node crashes, ignore it. Eg, 0 sends to 1: [5,6,0]
Wireless network: - step 1: 4 broadcasts to 6 and 8 - setp2: 6 broadcasts to everything it's connected to, so does 8. (build-tree phase) etc.. -reports best node
Large scale systems: - superpeers, each alloted certain nodes, low latency, evenly distributed - Repulsion of superpeers if not in its group
|
|
|
Sequential consistency: - if result of exec executed in same order. - appear in sequence and order p1: W(x)a _________________________ p2: W(x)b |
Causal consistency -Seen by all process -in same order
Concurrent writes - can be diff order on diff machines. i.e,read cannot be before write.
Grouping: Acq first, then Rel to read
Eventual consistency: replicated databases across WAN Monotonic read: if Pi reads x then p2 also reads same value of x or a more recent value. WS(x1;x2) - aka, write all the variables.
Read your writes: - effect of write on X will be seen by successive read on X
W(x1) -> R(x2)
Writes follow reads: Writes use last read/more recent version of X aka WS(x1;x2)
Replica server placement: cell size for placement
Conten replication: Permanent replica, Server-initiated replica, Client-initiated replica, Clients.
Remote write: primary server item takes all brunt Local write: Client2 server takes all the brunt(primary migrates to process wanting to perform update)
Quorum based: correct, conflict, ROWA |
