1 / 11

Towards_Self_Stabilizing_Wait_Free_Shared_Memory_Objects

“ Towards Self Stabilizing Wait Free Shared Memory Objects” By :  Hopeman  Tsigas  Paptriantafilou Presented By: Sumit Sukhramani Kent State University.

lane-mendez
Download Presentation

Towards_Self_Stabilizing_Wait_Free_Shared_Memory_Objects

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. “Towards Self Stabilizing Wait Free Shared Memory Objects”By:Hopeman Tsigas Paptriantafilou Presented By:Sumit SukhramaniKent State University

  2. Two types of faulty systems: Processors fail The memory is faultyProcessors fail: The distributed system must remain functional even if a certain fraction of the processors are malfunctioning. In the construction of shared memory objects like atomic objects, by considering wait free constructions we can guarantee any operation executed by a single processor to complete even if all other processors crash.Memory is faulty: A distributed system is required to overcome arbitrary changes to its state within a bounded amount of time. If the system does that then it is called self stabilizing.A shared memory object is a data structure stored in shared memory which may be accessed concurrently by several processors in the system through the invocation of operations for the object.

  3. As I proceed I’ll be covering the following topics: Defining self stabilized wait-free shared memory object. Wait free object of course will ensure that all operations after a transient error will eventually behave according to their specification. Next I’ll show that within this framework one cannot construct a self stabilizing single reader single writer regular bit from single reader single writer safe bits. Next we’ll see that how self stabilizing dual reader single writer safe bits can be used to construct single reader single writer regular bits.1. Defining Self stabilizing wait free objects: Shared registers are shared objects reminiscent of ordinary variables, but which can be read or written by different processors concurrently. Registers are constructed basically from the data structure of memory cells (sub registers) and set of read and write procedures. There are different levels of consistency in shared memory objects:a) Safe: A read returns most recently written value unless the read is concurrent with a write in which case returned values will be arbitrary.b) Regular: A read returns the value written by concurrent or immediately preceding write.c) Atomic: All operations appear to take effect instantaneously.

  4. LinearizabilityAlthoughoperations of concurrent processes may overlap, each operation appears to take effect instantaneously at some point between its invocation and response.The desired behavior of the object is described by its sequential specification S. This specifies the state if the object and for each operation its effect on the state of the object and its response. It is represented as (s, r = O (p), s| ) ε S if invoking O with parameters p in state s changes the state of the object to s|and returns r as its response.Definition1: An object is linearizable w.r.t. its sequential specification S if all possible runs over the object are linearizable w.r.t. S. This means that an object is linearizable w.r.t. specification S if all actions appear to take effect instantaneously and act according to S. Run over an object is basically a tuple <A, >with actions A and partial ordering  such that for A,B εA, AB iff the response for A occurred before the invocation of B. Runs thus have a infinite length and capture the real time ordering between actions invoked by the processors.Definition 2: A shared wait-free object is self stabilizing if an arbitrary fair execution ( in which all operations on all processors are executed infinitely often) started in an arbitrary state, is linearizable except for a finite prefix. Because this definition does not give a bound on the length of the finite prefix (which is stabilization time) and requires the processors to repair the transient fault, we come up with a stronger definition.Definition 3: A shared wait free object is self stabilizing if an arbitrary execution started in an arbitrary state is linearizable except for a bounded finite prefix.

  5. Definition 4: A run <A,> is linearizable w.r.t. sequential specification S after k processor actions, if there exists a corresponding sequential execution <A, > such that  is an extension of  and for all i if count (Ai) > k then (si, Ai, si+1) ε S.Here k is the delay of implementation. This definition gives the guarantee that all actions following arbitrarily behaving actions will agree on how that action actually behaved. For example all reads that occur immediately after a transient error should return the same value.Definition 5: An implementation of a shared object specified by sequential specification S is called k self stabilizing wait free if all its runs starting in an arbitrary initial state are linearizable w.r.t. S after k processor actions, and all actions complete within a bounded number of steps.In this definition the delay k of the implementation is taken to be independent of the type of operations performed by a processor.Definition 6: A register is k-stabilizing wait free if for all its runs starting in an arbitrary initial state only the first k actions preformed by one processor behave arbitrarily after the error, and all actions finish within a bounded number of steps. Such a register is safe if for all other read actions R not concurrent with a write action, R returns the value written by W with W   R. Such a register is regular if all other read actions return the value written by a feasible write.

  6. 0-Stabilizing 1W1R safe bits are not strong enoughConventionally it is taken that an adversary can chose a schedule, thereby forcing the protocols to behave incorrectly. The adversary is in control of: Scheduling the process steps in a run. Choosing the configuration of the system after a transient error.The writer (reader) does not have read access to the sub registers which it can write, in order to know its contents and to settle itself into correct stabilized behaviors, it has to rely on local information or on information passed by the shared registers that can be written by the reader. The adversary here can set the system in an inconsistent state. It thereby misleads the processes into incorrect behaviors.Theorem:There exists no deterministic implementation of a wait free k stabilizing 1W1R binary regular register using I W1R binary 0 stabilizing safe sub registers. Proof by contradiction

  7. Constructing a 0-Stabilizing 1W1R Regular BitWe construct here a 0-stabilizing 1W1R regular bit from a 0-stabilizing 1W2R safe bit.Similarity to Lamports construction of regular bit.For a binary registerthe only difference between regular and safe register is that if all feasible writes of a read operation write the same value a and at least one such write overlaps with the read, then for the safe bit the read may return either a or ¬a, whereas for a regular bit it must return a. But if the processor maintains a local copy of the last written value, and only writes a new value if this new value differs from the previous value stored in the register, then this is similar to a regular register. This approach although simple cannot be applied because as we have already seen that this approach cannot work for the implementation of a self stabilizing regular bit because due to a transient error the local copy of the last written value may be out of sync with the actual value present in the register.Theorem:The above protocol implements a wait free 0-stabilizing 1W1R regular binary register using one 0-stabilizing 1W2R safe binary register.

  8. Proof: We have to show that in <A,> all reads return the value written by a feasible write. Suppose a read RR returns the value a, since the register is binary this can be wrong only if all feasible writes write ¬a.  Then the write WR with WR   RR writes ¬a, and all writes WR| concurrent with RR write ¬a, hence we have WR   WR| .  WR and WR| read S without interference from <A|, | >. Either WR reads ¬a from S and hence does not write to S, or WR reads a from S and hence writes ¬a to S. Continuing this way we see that for all WR| concurrent with RR we see that none of them writes S. Hence, then RR reads S without interference in <A|, | > and as WR   RR reads ¬a from S and returns ¬a. This is contradiction.

  9. Constructing a 0-Stabilizing 1WIR l-Ary Regular RegisterHere we see the construction of a 0-stabilizing 1W1R l-ary regular register from l 0-stabilizing 1W1R regular bits. This protocol is also similar to regular multiple valued register by Lamport. The difference here is that we take into account that a transient error may cause all bits to be set to 0. In that case a default (l-1) is returned.

  10. Lemma: If WRi (1) |RRi then RRi returns 1 or there exists a j > I and a WR| (j) such that WR|j (1) | RRi+1. Moreover, if WRi (1) | RRi then WR (i)  Wr| (j).Theorem:Protocol 2 implements a 0-stabilizing 1W1R l-ary regular register using l 0-stabilizing 1W1R regular binary registers.Proof: This theorem can be proved by considering the definition 6 wherein we have to show that all reads return the value written by a feasible write. If for a read RR, for all I, RRi returns 0, then it returns l-1. Applying the above lemma inductively starting at i= l-1 we conclude that for no I, there is a WRi (1) such that WRi (1) |RRi .Constructing a 1-Stabilizing n WnR l-ary Atomic RegisterHere we construct a 1-stabilizing n WnR l-ary atomics register from 1R1W ∞ -ary o- stabilizing regular registers. In this protocol, V is the domain of values written and lead by the multi-writer register. The constructions use n2 regular 0-stabilizing regular registers Rij written by processor I and read by processor j. These register store a label consisting of an unbounded tag, a processor id with values in the domain {1,……,n} and a value in V. Labels are lexicographically ordered by ≤.

More Related