250 likes | 279 Views
Deadlock. Examples. P1 requests most of real memory Disk block mgr is swapped out to make room for P1's requests P1 requests disk block 1, which is held by p2 (& vice-versa for another block) Deadlock: disk block mgr cannot come in P1 cannot complete to get out. System Approaches.
E N D
Examples • P1 requests most of real memory • Disk block mgr is swapped out to make room for P1's requests • P1 requests disk block 1, which is held by p2 (& vice-versa for another block) • Deadlock: • disk block mgr cannot come in • P1 cannot complete to get out
System Approaches • Prevention • Avoidance • Detection & Recovery • Manual mgmt
Conditions for Deadlock • Mutual exclusion on R1 • Hold R1 & request on R2 • No preemption – once a R is requested, the request can't be retracted (because the app is now blocked! • Circularity • All 4 must apply simultaneously • Necessary, but NOT sufficient
Prevention • Design resource mgrs to always prevent at least ONE such condition • Easy in batch systems • Hard/impossible in other systems • Hard to apply to EVERY Rmgr
Avoidance • Predict effects of requests • refuse if deadlock could occur • Underutilizes R • Easy for batch systems • all requests are pre-defined
Detection & Recovery • Periodic (or manual) check for deadlock • implied by response time • expensive • non-productive until D fixed • D is indicated by non-occurrence • is it deadlocked or just waiting normally? • analysis of resource types (I/O vs code) • Recovery • preempt R from holder • delete offending process
Manual mgmt • Contemporary O/S's include detection & prevention algorithms • Not all R are covered due to cost (implementation or to users) • Often, simplest method is reboot
Resource State Diagrams A process P A resource R A request for R R is held by P
The State Transition Model • 3 possible events, E • request - ri • allocate - ai • deallocate - di • Pi P in sj S sk S due to x E sj sk x
A blocked process (P2) Circles are states, not processes. Subscripts represent processes.Arrows are transitions. a1 sj r3 r1 Transitions can occur OUT of Sjonly via the requests from P1 & P3or the allocation to P1.
Creating a complete state diagram Start with 1 process, P1, and only one R at a time may be requested. d d S0 S1 S2 S3 S4 r a r a Now duplicate this diagram for P2. Result is a complex diagram showing all possible states for P1 with all possible states for P2, as well as all the possible transitions.
Prevention : Hold & Wait • Must prevent holding followed by request • Two ways: • request everything at once • release all before making new requests
Prevention: Circular Wait • Draw system transition diagram or graph • Look for a prospective cycle • Disallow allocations that cause the cycle
Prevention: Allow preemption • Pn can "back-out" of a request • This is known as preempting the request State transitions e.g.; a move is possible from Sj to Sk when request n is honored rn sj sk dn rm sm Resource Request Graph
Avoidance • Similar to Prevention • Allows transition if guaranteed to be OK • Analyze new state before entering • System always safe • Unsafe state: no guarantee that deadlock won't occur
The Banker's Algorithm • maxc [ i, j ] is max claim for Rj by pi • alloc [ i, j ] is units of Rj held by pi • cj is the # of units of j in the whole system • Can always compute • avail [ j ] = cj - S0 i < nalloc [ i, j ] • and hence Rj available • Basically examine and enumerate all transitions Classic avoidance algorithm
Banker's Algorithm - Steps 1& 2 • // 4 resource types • C=# avail=<8, 5, 9, 7> • Compute units of R still available (C - col_sum) • avail [0] = 8 - 7 = 1 • avail [1] = 5 - 3 = 2 • avail [2] = 9 - 7 = 2 • avail [3] = 7 - 5 = 2 Step 1:allocalloc' Step 2:computations above yield: avail=<1,2,2,2> Current (safe) Allocation
Banker's Algorithm - Step 3 • Avail=<1,2,2,2> = # currently available for all Rj • Compute: maxc - alloc for each Pi (look for any satisfiable) • alloc' for P2 is <4,0,0,3> (from prev. table) • maxc[2, 0] - alloc'[2,0] = 5 - 4 = 1 ≤ avail[0] ≡ 1 • maxc[2, 1] - alloc'[2,1] = 1 - 0 = 1 ≤ avail[1] ≡ 2 • etc If no Pi satisfies: maxc - alloc' <= avail,then unsafe <stop> If alloc'=0 for all Pi <Stop> Maximum Claims
Banker's algorithm for P0 • maxc[0, 0] - alloc'[0,0] = 3 - 2 = 1 ≤ avail[0] ≡ 1 • maxc[0, 1] - alloc'[0,1] = 2 - 0 = 1 ≤ avail[1] ≡ 2 • maxc[0, 2] - alloc'[0,2] = 1 - 1 = 0 ≤ avail[2] ≡ 2 • maxc[0, 3] - alloc'[0,3] = 4 - 1 = 3 ≤ avail[3] ≡ 2 • Therefore P0 cannot make a transition to a safe state from the current state. • Likewise for P1
Banker's Algorithm - Step 4 • So P2 can claim, use and release all its Rigiving a new availability vector: avail2[0]=avail[0]+alloc'[2,0]=1+4=5 avail2[1]=avail[1]+alloc'[2,1]=2+0=2 avail2[2]=avail[2]+alloc'[2,2]=2+0=2 avail2[3]=avail[3]+alloc'[2,3]=2+3=5 avail2=<5,2,2,5> so at least one P can get its max claim satisfied
Serially Reusable Resources e.g.; disks, networks, CPUs P holds R (1 unit) A resource A process P requests R (1 unit) A resource A process
State Transitions due to Requests • In Sj, Pi is allowed to request qch units of Rh, provided pi has no outstanding requests. • Sj Sk, where the RRG for Sk is derived from Sj by adding q request edges from pi to Rh q edges pi Rh pi Rh pi requests q units State Sk State Sj of Rh RRG=Resource Request Graph
Transitions P0 P0 • subscript on state indicates who the requestor was. • r1 is a transition: request for the resource by P1 r1 P1 P1 s00 s01
Consumable Resource Graphs P produces R P requests R (1 unit) P requests R (2 unit)