Surveyor/Respondent Scalability Protocol

A fairly common problem in building distributed applications is peer discovery -- or how do you find your peers. For example, imagine an internet chat type application, where server wants to determine the presence of all peers, including perhaps some information such as their unique social networking handle. Another similar problem involves voting algorithms, where a survey of all connected peers is required to arrive to some solution to a problem. This is common with distributed consensus algorithms. One of the most common problems in distributed applications is how to delegate a work to another processing node and get the result back to the original node. In other words, the goal is to utilise the CPU power of a remote node. It turns out that these problems are very similar. We can assume potential participants will register with a central process. Once that is done, the central process can send out a survey request to the participants when it wants to perform a survey. Also, note that it is reasonable and possible for a participant to decline to participate (i.e. decline to respond.) This can happen due to loss of network connectivity, or can represent a conscious decision on the part of the respondent. For example, a real-world example of this would be asking audience members to raise their hands if they like the color red. The act of raising one's hand can be thought of as responding. As a consequence, taken generally, the surveyor should not infer any thing about parties it doesn't get a response from. Perhaps the respondent simply didn't hear the question, or perhaps she declines to self-identify. This measn that surveying should be thought of as a best-effort service. Applications which need more resilience may repeat their inquiries. It is common in other networking protocols to do so periodically, and only "expire" the response from a peer that is non-responsive after it has missed several successive surveys. Furthermore, the act of asking a question has to be time bounded. This is particularly important if multiple surveys are to be issued. Sufficient time for responses from the first survey to occur must pass before starting a new one, unless some other identifying content is present to distinguish the results from one survey from another. (Going back to our raised hands, imagine two questions asked in rapid succession, one if you like the color red, the other if you like the color blue. If only one hand is used, and there is not sufficient time between the questions, it becomes impossible to distinguish which color is preferred. Of course, if one uses two hands -- a distinguishing identifier, now we can have two surveys running in parallel. Fortunately we usually have more bits available for conveying this kind of information in network protocols.) In all cases the act of surveying and replying can be thought of as state-less. In otherwords, a given response should not depend upon the content of any prior surveys. Ideally, because of the best-effort nature of this, it is also beneficial if surveying is itself idempotent, i.e. the act of responding to a survey should not itself change state on the respondent. Generally there are few common scenarios that come up with real-world situations. Here are some of them. One surveyor issues one survey, and then zero, one or many responders reply. The surveyor collects then these responses over a period of time before issuing a new survey. One surveyor issues multiple surveys, distinguishing which replies are to which survey based on some identifying content. For example, this can be thought of like ARP, where multiple requests can be outstanding. Multiple surveyors issue surveys, but one each at a time. Responders reply to each of these as appropriate. For example, imagine a network with two print clients and a number of networked printers. Both clients may occasionally desire to inquire as supply levels, and since they don't talk to each other, the replies may go to either system. Multiple surveyors issuing multiple surveys concurrently. This is the combination of the second and third cases above.

The surveyor/respondent protocol can be run on top of any SP mapping, such as, for example, SP TCPmapping. Also, given that SP protocols describe the behaviour of entire arbitrarily complex topology rather than of a single node-to-node communication, several underlying protocols can be used in parallel. For example, a client may send a request via WebSocket, then, on the edge of the company network an intermediary node may retransmit it using TCP etc.

+---+ WebSocket +---+ TCP +---+ | |-------------| |-----------| | +---+ +---+ +---+ | | +---+ IPC | | SCTP +---+ DCCP +---+ | |---------+ +--------| |-----------| | +---+ +---+ +---+

Surveyor/respondent protocol defines two different endpoint types: The SURVEYOR and the replier or RESPONDENT. A SURVEYOR endpoint can be connected only to a RESPONDENT endpoint, and vice versa. If the underlying protocol indicates that there's an attempt to create a channel to an incompatible endpoint, the channel MUST NOT be used. In the case of TCP mapping, for example, the underlying TCP connection MUST be closed. When creating more complex topologies, SURVEYOR and RESPONDENT endpoints are paired in the intermediate nodes to form a forwarding component, so called "device". Device receives requests from the SURVEYOR endpoint and forwards them to the RESPONDENT endpoint. At the same time it receives replies from the RESPONDENT endpoint and forwards them to the SURVEYOR endpoint:

--- surveys --> +----------+ +------------+----------+ +------------+ | |-->| | |-->| | | SURVEYOR | | RESPONDENT | SURVEYOR | | RESPONDENT | | |<--| | |<--| | +----------+ +------------+----------+ +------------+ <-- responses --- Using devices, arbitrary complex topologies can be built. The rest of this section explains how are the requests routed through a topology towards processing nodes and how are responses routed back from processing nodes to the original clients. Because the delivery of both surveys and responses is handled on a best-effort basis, when the transport is faced with pushback, it is acceptable for the implementation to drop the message. Applications expecting resilience in the face of such events should expect to perform multiple surveys over time; a failure to respond to a survey shall not be taken as a critical fault. As for delivering replies back to the clients, it should be understood that the client may not be directly accessible (say using TCP/IP) from the processing node. It may be beyond a firewall, have no static IP address etc. Furthermore, the client and the processing may not even speak the same transport protocol -- imagine client connecting to the topology using WebSockets and processing node via SCTP. Given the above, it becomes obvious that the replies must be routed back through the existing topology rather than directly. In fact, surveyor/respondent topology may be thought of as an overlay network on the top of underlying transport mechanisms. As for routing replies within the surveyor/respondent topology, it is designed in such a way that each reply contains the whole routing path, rather than containing just the address of destination node, as is the case with, for example, TCP/IP. The downside of the design is that surveys and responses are a little bit longer. Also this assumes symmetric connectivity in the underlying transports. The upside, on the other hand, is that the nodes in the topology don't have to maintain any routing tables beside the simple table of adjacent channels along with their IDs. There's also no need for any additional protocols for distributing routing information within the topology. The most important reason for adopting the design though is that there's no propagation delay and any nodes becomes accessible immediately after it is started. Given that some nodes in the topology may be extremely short-lived this is a crucial requirement. Imagine a database client that sends a survey, gets a single response, and then immediately answers. (Think of a simple question like "is anyone here?" A single reply is sufficies to answer the question.) It makes no sense to delay the whole process until the routing tables are synchronised between the client and the server. The algorithm thus works as follows: When a survey is routed from the client to the processing node, every RESPONDENT endpoint determines which channel it was received from and adds the ID of the channel to the survey. Thus, when the survey arrives at the ultimate respondent it already contains a full backtrace stack, which in turn contains all the info needed to route a message back to the original surveyor. After processing the survey, the responding node attaches the backtrace stack from the survey to the response and sends it back to the topology. At that point every RESPONDENT endpoint can check the traceback and determine which channel it should send the reply to. In addition to routing, surveyor/respondent protocol takes care of matching responses and surveys. That is, it can ensure that a given response cannot be mismatched to a different survey. In order to avoid confusion, after the surveyor has received all the responses it expects to (typically when a period of time has passed), it should discard further stray responses. The surveyor thus adds an unique request ID to the survey. The ID gets copied from the survey to the response by the responding node. When the response gets back to the surveyor, it can simply check whether the survey in question is still being outstanding and if not so, it can ignore the response. To implement all the functionality described above, messages (both surveys and responses have the following format:

+-+------------+-+------------+ +-+------------+-------------+ |0| Channel ID |0| Channel ID |...|1| Request ID | payload | +-+------------+-+------------+ +-+------------+ ------------+ The payload of the message is preceded by a stack of 32-bit tags. The most significant bit of each tag is set to 0 except for the very last tag. That allows the algorithm to find out where the tags end and where the message payload begins. As for the remaining 31 bits, they are either survey ID (in the last tag) or a channel ID (in all the remaining tags). The first channel ID is added and processed by the RESPONDENT endpoint closest to the processing node. The last channel ID is added and processed by the RESPONDENT endpoint closest to the client. Following picture shows an example of request saying "Hello" being routed from the client through two intermediate nodes to the processing node and the reply "World" being routed back. It shows what messages are passed over the network at each step of the process:

client Hello | World | +------------+ ^ | | SURVEYOR | | V +------------+ | 1|823|Hello | 1|823|World | +------------+ ^ | | RESPONDENT | | | +------------+ | | | SURVEYOR | | V +------------+ | 0|299|1|823|Hello | 0|299|1|823|World | +------------+ ^ | | RESPONDENT | | | +------------+ | | | SURVEYOR | | V +------------+ | 0|446|0|299|1|823|Hello | 0|446|0|299|1|823|World | +------------+ ^ | | RESPONDENT | | V +------------+ | Hello | World server

All endpoints implement so called "hop-by-hop" functionality. It's the functionality concerned with sending messages to the immediately adjacent components and receiving messages from them. To make an analogy with the TCP/IP stack, IP provides hop-by-hop functionality, i.e. routing of the packets to the adjacent node, while TCP implements end-to-end functionality such resending of lost packets. As a rule of thumb, raw hop-by-hop endpoints are used to build devices (intermediary nodes in the topology) while end-to-end endpoints are used directly by the applications. To prevent confusion, the specification of the endpoint behaviour below will discuss hop-by-hop and end end-to-end functionality in separate chapters.

The SURVEYOR endpoint is used by the user to send surveyor to the responding nodes and receive the responses afterwards. When user asks the SURVEYOR endpoint to send a request, the endpoint should send it to ALL of the associated outbound channels (TCP connections or similar). The request sent is exactly the message supplied by the user. SURVEYOR sockets MUST NOT modify an outgoing survey in any way. If there's no channel to send the survey to, the survey is merely discarded. The endpoint MAY report the backpressure condition to the user as well. If there are associated channels but none of them is available for sending, i.e. all of them are already reporting backpressure, the endpoint won't send the message and MAY report the backpressure condition to the user. The actual survey is discarded. If the channel is not capable of reporting backpressure (e.g. DCCP) the endpoint SHOULD consider it as always available for sending new request. When there are multiple channels available for sending the survey endpoint MUST deliver the survey to all of them. As for incoming messages, i.e. responses, SURVEYOR endpoints MUST fair-queue them. In other words, if there are replies available on several channels, they MUST receive them in a round-robin fashion. They must also take care not to compromise the fairness when new channels are added or old ones removed. In addition to providing basic fairness, the goal of fair-queueing is to prevent DoS attacks where a huge stream of fake responses from one channel would be able to block the real replies coming from different channels. Fair queueing ensures that messages from every channel are received at approximately the same rate. That way, DoS attack can slow down the system but it can't entirely block it. Incoming responses MUST be handed to the user exactly as they were received. SURVEYOR endpoints MUST not modify the responses in any way.

RESPONDENT endpoints are used to receive surveys from the clients and send resopnses back to the clients. First of all, each RESPONDENT socket is responsible for assigning unique 31-bit channel IDs to the individual associated channels. The first ID assigned MUST be random. Next is computed by adding 1 to the previous one with potential overflow to 0. The implementation MUST ensure that the random number is different each time the endpoint is re-started, the process that contains it is restarted or similar. So, for example, using pseudo-random generator with a constant seed won't do. The goal of the algorithm is to the spread of possible channel ID values and thus minimise the chance that a response is routed to an unrelated channel, even in the face of intermediate node failures. When receiving a message, RESPONDENT endpoints MUST fair-queue among the channels available for receiving. In other words they should round-robin among such channels and receive one request from a channel at a time. They MUST also implement the round-robin algorithm is such a way that adding or removing channels doesn't break the fairness. In addition to guaranteeing basic fairness in access to computing resources the above algorithm makes it impossible for a malevolent or misbehaving client to completely block the processing of requests from other clients by issuing steady stream of surveys. After receiving the survey, the RESPONDENT socket should prepend it by 32 bit value, consisting of 1 bit set to 0 followed by the 31-bit ID of the channel the request was received from. The extended survey will be then handed to the user. The goal of adding the channel ID to the response is to be able to route the response back to the original channel later on. Thus, when the user sends a response, endpoint strips first 32 bits off and uses the value to determine where it is to be routed. If the response is shorter than 32 bits, it is malformed and the endpoint MUST ignore it. Also, if the most relevant bit of the 32-bit value isn't set to 0, the response is malformed and MUST be ignored. Otherwise, the endpoint checks whether its table of associated channels contains the channel with a corresponding ID. If so, it sends the response (with first 32 bits stripped off) to that channel. If the channel is not found, the response MUST be dropped. If the channel is not available for sending, i.e. it is applying backpressure, the response MUST be dropped. Note that when the response is unroutable two things might have happened. Either there was some kind of network disruption, in which case the survey may be re-sent later on, or the original client have failed or been shut down. In such case the survey won't be resent, however, it doesn't really matter because there's no one to deliver the response to any more anyway. Unlike surveys, there's never pushback applied to the responses; they are simply dropped. If the endpoint blocked and waited for the channel to become available, all the subsequent replies, possibly destined for different unblocked channels, would be blocked in the meantime. That allows for a DoS attack simply by firing a lot of surveys and not receiving the responses.

End-to-end functionality is built on top of hop-to-hop functionality. Thus, an endpoint on the edge of a topology contains all the hop-by-hop functionality, but also implements additional functionality of its own. This end-to-end functionality acts basically as a user of the underlying hop-by-hop functionality.

End-to-end functionality for SURVEYOR sockets is concerned with matching the responses to surveys, and with filtering out stray or outdated responses. To be able to do this, the endpoint must tag the survey with unique 31-bit survey IDs. First survey ID is picked at random. All subsequent survey IDs are generated by adding 1 to the last survey ID and possibly overflowing to 0. To improve robustness of the system, the implementation MUST ensure that the random number is different each time the endpoint, the process or the machine is restarted. Pseudo-random generator with fixed seed won't do. When user asks the endpoint to send a message, the endpoint prepends a 32-bit value to the message, consisting of a single bit set to 1 followed by a 31-bit survey ID and passes it on in a standard hop-by-hop way. If the hop-by-hop layer reports pushback condition, the end-to-end layer considers the survey unsent and MAY report pushback condition to the user. If the survey is successfully sent, the endpoint stores the survey including its survey ID, so that it can be resent later on if needed. At the same time it sets up a timer to receive all of the responses. The user MUST be allowed to specify the timeout interval. The default timeout interval must be 60 seconds. When a response is received from the underlying hop-by-hop implementation, the endpoint should strip off first 32 bits from the response to check whether it is a valid reply. If the response is shorter than 32 bits, it is malformed and the endpoint MUST ignore it. If the most significant bit of the 32-bit value is set to 0, the reply is malformed and MUST be ignored. Otherwise, the endpoint should check whether the survey ID in the response matches any of the survey IDs of the surveys being processed at the moment. If not so, the response MUST be ignored. It is either a stray message or a too-long delayed response. Please note that the endpoint can support either one or more surveys being processed in parallel. Which one is the case depends on the API exposed to the user and is not part of this specification. If the ID in the response matches one of the surveys in progress, the response MUST be passed to the user (with the 32-bit prefix stripped off). A SURVEYOR endpoint MUST make it possible for the user to cancel a particular survey in progress. What it means technically is deleting the stored copy of the survey and cancelling the associated timer. Thus, once the response arrives, it will be discarded by the algorithm above. Finally, when the timeout for a survey expires, then the survey must be canceled in a manner similar to user-initiated cancelation. That is, the stored copy of the survey must be deleted, the timer removed, and any further responses received with the same survey ID are subsequently discarded.

End-to-end functionality for RESPONDENT endpoints is concerned with turning surveys into corresponding responses. When user asks to receive a survey, the endpoint gets next request from the hop-by-hop layer and splits it into the traceback stack and the message payload itself. The traceback stack is stored and the payload is returned to the user. The algorithm for splitting the survey is as follows: Strip 32 bit tags from the message in one-by-one manner. Once the most significant bit of the tag is set, we've reached the bottom of the traceback stack and the splitting is done. If the end of the message is reached without finding the bottom of the stack, the survey is malformed and MUST be ignored. Note that the payload produced by this procedure is the same as the survey payload sent by the original client. Once the user processes the survey and sends the response, the endpoint prepends the response with the stored traceback stack and sends it on using the hop-by-hop layer. At that point the stored traceback stack MUST be deallocated. Additionally, RESPONDENT endpoints MUST support cancelling any survey being processed at the moment. What it means, technically, is that state associated with the survey, i.e. the traceback stack stored by the endpoint is deleted and reply to that particular survey is never sent. The most important use of cancellation is allowing the service instances to ignore surveys (whether due to malformation or for other application specific reasons.) In such case the reply is never sent. Of course, if application wants to send an application-specific error massage back to the client it can do so by not cancelling the survey and sending a regular response.

It may happen that a request/reply topology contains a loop. It becomes increasingly likely as the topology grows out of scope of a single organisation and there are multiple administrators involved in maintaining it. Unfortunate interaction between two perfectly legitimate setups can cause loop to be created. With no additional guards against the loops, it's likely that requests will be caught inside the loop, rotating there forever, each message gradually growing in size as new prefixes are added to it by each RESPONDENT endpoint on the way. Eventually, a loop can cause congestion and bring the whole system to a halt. To deal with the problem SURVEYOR endpoints MUST check the depth of the traceback stack for every outgoing request and discard any requests where it exceeds certain threshold. The threshold SHOULD be defined by the user. The default value is suggested to be 8.

New SP endpoint types SURVEYOR and RESPONDENT should be registered by IANA. For now, value of 98 should be used for SURVEYOR endpoints and value of 99 for RESPONDENT endpoints. (An earlier similar protocol without the backtrace headers used protocol numbers 96 and 97.)

The mapping is not intended to provide any additional security to the underlying protocol. DoS concerns are addressed within the specification.