Asynchronous checkpointing and recovery : Juang-Venkatesan algorithm Coming to the next type of checkpointing and recovery algorithm, which is said to be asynchronous checkpointing and recovery. Right, asynchronous checkpointing and recovery was proposed by Juang-Venkatesan, so it’s also called as Juang-Venkatesan algorithm. Here, there are some assumptions. As usual, the communication channels are reliable, delivery of message in FIFO order, infinite buffering, message transmission delay is arbitrary but not finite. But finite, right? It is arbitrary, meaning the delay will happen sometimes, the delay will be short, and the delay will be long. As it happens at one particular time and happens at another particular time, we cannot say it is arbitrary, but it is finite. Finite number of delays only happen. So, these and all are assumptions made by this algorithm, Juang-Venkatesan algorithm. Here, there are some assumptions. As usual, the communication channels are reliable, delivery of message in FIFO order, infinite buffering, message transmission delay is arbitrary but not finite. But finite, right? It is arbitrary, means delay will happen sometimes, the delay will be short, and delay will be long, as it happens at one particular time and happens at another particular time. We cannot, uh, say it is arbitrary, but it is finite. Finite number of delay only happens. So, these and all are assumptions made by this algorithm, John Vincation algorithm. Okay, and the compute, the underlying computation, otherwise the application is event-driven, right? That means the process pays at status, receives message m, processes the message, moves to state s dash. Okay, so this is what event-driven. So, process is changing state to state. If it’s yes, it’s in yes, sending message, after that, it becomes s dash. Okay? Okay, so this is what actually, uh, the event-driven. So that the process, every process, is denoted as a triplet here, the triplet s, comma, m, comma, messages end, which is, which represents the state of any process p. Okay, so here, yes, say yes denotes the state, m denotes the message, and message is sent to denotes the content of the message. And two types of log storage are maintained. One is volatile log. Volatile means we already know it will vanish after some time. Okay, so this is a short time to access, but lost if processor crashes, right? And if at all it moved to the stable log periodically. Okay, it’s a temporary login, and there is a stable log also, which is kept for longer time, even though the processor gets crashed, right? Okay, so these are the basics you have to know before getting into the John Vegetation algorithm. And here, the checkpointing is asynchronous, right? Asynchronous means there is no cooperation. That is, in the previous case, it’s synchronous because all the processes are cooperating for a time, right? They will be obedient and take decision at a time, isn’t so? By that decision-taking duration, they won’t be allowed to send any message or something else. It is for our common room meeting, nothing, right? But asynchronous checkpointing is not like that. Asynchronously, the processors are not being controlled, right? So, after executing an event, the triplet is recorded, right? Without any synchronization with other processes. That means, a process, whenever and whenever it nessa is, it is free or otherwise arbitrarily they may lock the triplet, right? And there is a local checkpoint consists of a set of records, right? First star stored in volatility, then move to stable storage. So, let us suppose to learn the notations and idea used behind the recovery. Here, the notation called rcvd, which is received in the padilla, received of i from j, received i from j, right? Where j, at the point of checkpoint, local checkpoint, ckpti, is the local checkpoint of the process i, right? So, received of i, process i, from j, at the local checkpoint will be the number of messages received by pi from pj, from the beginning of the computation of the checkpoint of ckpti. Okay, and what is sent? This message is sent, i to j, ckpti, at the time of ckpti, is referred to the number of messages sent by pi to pj. That is very, very important. Either confusion will come out from pa, from tj, you reverse a pa to fiji, right? You received a representative. Okay, there are two processes involved here, pi and pj. Okay, so here, sender and received. Okay, this is what the scenario, right? So, the number of messages sent by pi to pj from the beginning of the computation at checkpoint ckpti, right? At a particular checkpoint, how many messages are received from pj to pi? Similarly, how many messages are sent to pj from pi? Right? So, this is what the notation, and another idea. Idea here is, from the set of checkpoints, find consistent among them. Find a set of consistent points. Doing that, based on the number of messages sent and received, you will have an idea. Here, the procedure for rollback recovery. Processor pa executes the following, right? So, pi processor pi executes first is step pa. This is the step a, and this is the step b. During step a, what happens is, if processor pa is recovering from a failure, processor pa happens to fail, it recovers from its failure, then what it does is see it. It takes the local checkpoint ck pti. Okay, so this ck pti will be obtained by referring to the latest event because there is a log used to know stable storage, right? So, stable storage only can be able to serve because it is permanent. Volatile storage, once after the crash occurs, it vanishes. So, here the log details are available in stable storage. So, with referring the stable storage log, the process pa can obtain the checkpoint where that is before, uh, crashed. What was the state ckr, ck pti, that is what ck pti. Otherwise, yes, ck pti will be okay. Suppose the processor pa recovering after failure, stable storage, otherwise, even that took place in pa the very latest event. Okay, it is either in stable storage or in volatile, because that doesn’t happen, layer failure doesn’t happen in this case, so volatile storage, right? So, there is a process pi here, failure occurs. Failure occurs, so at this particular stage, the even checkpoint will be logged into, will be logged into storage. So, once after failure, when a crash occurs and recovery happens, here the process comes and refers this pti, is learnt, it comes to know what is ckpti, then it says the process gets into step b. Instead b, for k is equal to 1 to n, where n is the number of other processors involved in the system, right? Okay, for example, there are three processes, p one, p two, p three, for d naught, n is equal to three. Okay, for every process, for each neighboring process, every other process, for example, p i start within, ah, for example, there are three processes, if p1 start, it takes p2 and p3 also, for each neighboring processor pj do computing how many messages are sent from the checkpoint ck pti, right? So, for example, in this case, if suppose p1 starts, okay, if so, if there are so other processes like p2, p3, now p1 starts, suppose p1 sends message here and here, now this, this is the ck pti, right? And, uh, here p2 have sent already a message, yeah, this is the scenario, p1 has sent two messages, p2 has sent one message. Right now, computing send off, sent off i to j after ck pti, how many messages here? 2 i 2 j means here p1 is i, p2 is j, right? So, because we are talking from p1, so all the neighboring nodes here, the pi here is the k here is 1, so it is p1 for other processor two at the ckpti. How much it is? Two, then sent off one to three at ckpti equal to zero, because there is no message transfer between p1 and p3, right? Okay, send a rollback after computing this send command, it then send a rollback message, right? Rollback off i, comma sent off to jp, take a message to pj, right? In the morning, sorry, message. So, this happens, so for the first time, this for loop is executing for each neighboring processor. First, the neighboring process is taken as p2, send a rollback next to the neighboring processor, taken as p3, send a send, send, calculate send and rollback. Now, the algorithm, yes, in the for loop in diagram, the murray send rollback when in the for loop end item, right? Now, what happens is, for every rollback, for every rollback j, comma c, right? Right message received from a neighbor j, right? Back off j, comma c, right? Sip in the case rollback off i, comma sent off i to j, c k p t i, know off. What is i, which is i is one, p1, layer that’s what i is one. What is center phi to j, c k p t i, here center of i to j, one to show, tk p h two, so one comma two, rollback off 1 comma 2. Okay, so p1 sends a rollback message of 1 comma 2 to p2. Okay, so that’s what the rollback message or the content is. Now, for every rollback j, comma c, arena rollback j, comma c, this is j, this is c, right? Received from the neighbor j, at the receiver end, what happens if received of ckpti for sensible, here is j, which is p2, at the p2 ending a p1 and delay p2 and it calculate received message, what is rc vdf received of i from j, j to i, layer? What is received of j to i at the point c k pti? At point c k pti, j received two messages, isn’t so two messages, so it is two, right? Just happened, yep. Received a j from, sorry, j to i, sorry, received off i to j is ck pti is greater than c, so it seems to be an open message, seems to be an open message, right? For example, before this checkpoint only two messages sent after checkpoint one and appear ground, but at the count of p2 it is three, so these two messages only p1 remembers and sends rollback of one comma two, but for this message there is nothing calculated because this message, the parent is not accounting since it is rule recovered. So, that always here, this condition satisfied, received of ckpti is more than c, right? That is the rollback solar panel india message and message, right? Then find the latest event e such that received of e equal to c, okay? Community even t may be the volatile storage or so, maybe in the volatile storage or in the stable storage. So, by taking that previous e, the checkpoint of the jth process will be changed. The checkpoint of this jth process is changed. The processor understand that there is an open message and finds the nearest checkpoint at this previous checkpoint, which is called the event e. Let us even t with refer to the stable storage or volatile storage, so the checkpoint of the process, second process is rewinded to the previous location, right? So, this is how actually for p1 process k is equal to 2, now p2 starts, p2 starts sending message. How many it is it has sent to p1 and p3 similarly it has to roll back, then p1 and p3 are checking whether it is right or wrong, if it is right roll right, so this is so actually up and then once completely it is over then p3, the turn comes to p3, okay? So, this is all actually rollback recovery algorithm works for example.