For MQTT Smarter is Better
For MQTT smarter is better: MQTT is the protocol of choice for many industrial communication tasks, but was not developed for IIoT in mind.
For MQTT smarter is better: MQTT is the protocol of choice for many industrial communication tasks, but was not developed for IIoT in mind.
Accessing production data vs cybersecurity? Why not both? This white paper explains how you can have it both ways.
Easily access OPC A&E from multiple network sources, or convert it to OPC DA, UA and other protocols using DataHub middleware.
An enhanced, secure-by-design OPC UA to MQTT gateway can pass data through a DMZ or IT department, keeping all inbound firewall ports on the plant closed.
Agood IIoT protocol is the basis for effective IIoT data communication. Without a secure, robust IIoT protocol, data can be late, missing, inconsistent, or dangerously incorrect, leading to costly errors and wasted time.
With the IIoT still in its infancy, companies have turned first to familiar, well-tested data communication and messaging protocols such as MQTT, AMQP, REST and OPC UA for an IIoT protocol. Valid as these may be for their designed purposes, they were never intended to support IIoT data communication. Thus, when evaluated according to criteria for a robust, secure Industrial IoT implementation, they all come up somewhat short.
Skkynet’s software and services are designed for the IIoT, and meet all of the criteria for effective data communication. Here we provide a comparison report on how well MQTT, AMQP, REST, OPC UA, and Skkynet’s own DHTP (DataHub Transfer Protocol) meet the criteria summarized in the above table for an ideal IIoT protocol. Each of the criteria enumerated above is explained in further detail in subsequent sections.
Keeps all inbound firewall ports closed for both data sources and data users.
Keeping all inbound firewall ports closed at the plant resolves many security issues for Industrial IoT. MQTT, AMQP, REST and DHTP meet this criterion. OPC UA does not because it has a client/server architecture, which requires at least one firewall port be open on the server side (typically the plant) to allow for incoming client connections. This is an unacceptable risk for most industrial systems. Skkynet’s DataHub and ETK connect locally to servers and clients in the plant, and make outbound connections via DHTP to SkkyHub running on a cloud server, or to another DataHub running on a DMZ computer. This outbound connection keeps all inbound firewall ports closed and hides the plant from the outside world.
Consumes minimal bandwidth, while functioning with the lowest possible latency.
One goal of any industrial communication or IIoT protocol is to consume as little bandwidth as possible, and function with the lowest possible latency. MQTT and AMQP do this well. REST does not, because every transaction includes all of the socket set-up time and communication overhead. OPC-UA is partial, because it uses a smart polling mechanism that trades bandwidth for latency. Skkynet software and services maintain a connection and transmit only the data via DHTP, consuming very little bandwidth, at very low latencies.
Can support hundreds or thousands of interconnected data sources and users.
An important aspect of the Internet of Things is the vision of connecting hundreds, thousands, and even millions of things via the Internet, and providing access to the data from any single thing, or groups of things to any number of clients. Event-driven protocols like MQTT and AMQP allow for this kind of scaling up, while REST’s polling model prevents it. OPC UA is also event-driven, and so theoretically can scale up, but its underlying polling model does not allow for very large numbers of simultaneous connections. DHTP abstracts the data from the protocol across the connection, and also implements an event-driven model, which allows it to scale up well.
Adds virtually no latency to the data transmission.
Any kind of remote HMI or supervisory control system is much more effective when functioning in at least near-real time. Propagation delays of one or more seconds may be tolerable under certain conditions or for certain use cases, but they are not ideal. AMQP and MQTT offer real-time behavior only if they are not operating with a delivery guarantee. That is, if you choose the “guaranteed delivery” quality of service then a slow connection will fall further and further behind real-time. By contrast, DHTP guarantees consistency, not individual packet delivery, and can sustain that guarantee in real time on a slow connection. REST simply has too much connection overhead to allow real-time performance in most circumstances. OPC UA, being an industrial protocol, meets this criterion well.
Encodes the data so that clients and servers do not need to know each other’s protocols.
A well-defined data format is essential for interoperability, allowing any data source to communicate seamlessly with any data user. Interoperability was the primary driving force behind the original OPC protocols, and is fully supported by the OPC UA data format. Any Industrial IoT software or service should support at least one, if not multiple interoperable data formats. Skkynet’s DataHub software and ETK support several, and allow for real-time interchange between them and DHTP. MQTT, AMQP and REST do not support interoperability between servers and clients because they do not define the data format, only the message envelope format. Thus, one vendor’s MQTT server will most likely not be able to communicate with another vendor’s MQTT client, and the same is true for AMQP and REST.
A messaging broker responds appropriately when a data user is unable to keep up with the incoming data rate.
Overload handling refers to how the broker responds when a client is unable to keep up with the incoming data rate, or when the server is unable to keep up with the incoming data rate from the client. MQTT and AMQP respond in one of two ways. Either they block, effectively becoming inoperative and blocking all clients. Or they drop new data in favor of old data, which leads to inconsistency between client and server. REST saturates its web server and becomes unresponsive. OPC UA attempts to drop old data in favor of new data, but consumes massive amounts of CPU resources to do so. When needed, Skkynet’s DataHub and SkkyHub can drop old data efficiently, and using DHTP they guarantee consistency between client and server even over multiple hops. Data coming from or going to overloaded clients remains consistent, and all other clients are unaffected.
Each client application knows with certainty if and when a connection anywhere along the data path has been lost, and when it recovers.
Most protocols do not provide failure notification information from within the protocol itself, but rather rely on clients to identify that a socket connection is lost. This mechanism does not propagate when there is more than one hop in the communication chain. Some protocols (such as MQTT) use a “last will and testament” that is application-specific and thus not portable, and which is only good for one connection in the chain. Clients getting data from multiple sources would need to be specifically configured to know which “last will” message is associated with which data source. In MQTT, AMQP, REST and OPC UA alike, the protocol assumes that the client will know how many hops the data is traversing, and that the client will attempt to monitor the health of all hops. That is exceptionally fragile, since knowledge about the data routing must be encoded in the client. In general, this cannot be made reliable. DHTP propagates not only the data itself, but information about the quality of the connection. Each node is fully aware of the quality of the data, and passes that information along to the next node or client.
Guarantees consistency of data, preserved through multiple hops.
An important goal of the IIoT is to provide a consistent picture of the industrial data set, whether for archival, monitoring, or supervisory control. MQTT’s ability to guarantee consistency of data is fragile because its Quality of Service options only apply to a single hop in the data chain. And within that single hop, delivery can be guaranteed only at the expense of losing real-time performance. Real-time performance can be preserved, but only by dropping messages and allowing data to become inconsistent between client and server. AMQP’s ability to guarantee consistency of data is fragile because like MQTT it only applies to a single hop in the chain. Additionally, its delivery guarantee blocks when the client cannot keep up with the server and becomes saturated. REST provides no Quality of Service option, and while OPC UA guarantees consistency it cannot work over multiple hops. DHTP guarantees consistency, and the guarantee is preserved through any number of hops.
Brokers can connect to other brokers to support a wide range of collection and distribution architectures.
The requirements of the IIoT take it beyond the basic client-to-server architecture of traditional industrial applications. To get data out of a plant and into another plant, corporate office, web page or client location, often through a DMZ or cloud server, typically requires two or more servers, chained together. The OPC UA protocol is simply too complex to reproduce in a daisy chain. Information will be lost in the first hop. Attempts to daisy chain some aspects of the OPC UA protocol would result in synchronous multi-hop interactions that would be fragile on all but the most reliable networks, and would result in high latencies. Nor would OPC UA chains provide access to the data at each node in the chain. REST servers could in theory be daisy chained, but would be synchronous, and not provide access to the data at each node in the chain. MQTT and AMQP can be chained, but it requires each node in the chain to be aware that it is part of the chain, and to be individually configured. The QoS guarantees in MQTT and AMQP cannot propagate through the chain, so daisy chaining makes data at the ends unreliable. Skkynet’s DataHub and SkkyHub both support daisy-chained servers because DHTP allows them to mirror the full data set at each node, and provide access to that data both to qualified clients, as well as the next node in the chain. The DHTP QoS guarantee states that any client or intermediate point in the chain will be consistent with the original source, even if some events must be dropped to accommodate limited bandwidth.
Far from exhaustive, this overview of effective IIoT data communication provides an introduction to the subject, and attempts to highlight some of the key concepts, through sharing what we have found to be essential criteria for evaluating some of the protocols currently on offer. Because none of MQTT, AMQP, REST, or OPC UA were designed specifically for use in Industrial IoT, it is not surprising that they do not fulfill these criteria. DHTP, on the other hand, was created specifically to meet the needs of effective industrial and IIoT data communication, making it an ideal choice for an IIoT protocol.
Note: This article was originally published on the Automation.com website.
OPC UA was designed to be secure in an industrial environment, and it does a good job of it. In the world of Operations Technology (OT) you need reliable and secure data communications to run mission-critical systems. OPC UA provides robust connectivity, allowing your devices and machines to communicate, yet keeping them secure and locked down. But today’s OT world is expanding, being propelled into the larger, corporate world of IT, and beyond that, into the Industrial Internet of Things (IIoT) and Industrie 4.0. When connecting to IT and the IIoT, making OPC UA secure requires a new approach to meet new and different threats to security.
Securing an industrial system requires at the very least securing the perimeter against unauthorized access. Whether or not anything in the plant is connected to IT or the IIoT, this perimeter must remain intact for optimal security. In the past, perimeter protection was often accomplished by air-gapping, where the industrial network was physically isolated from any other network connection. Until recently, this approach or similar solutions like DMZs were sufficient. But these make it difficult if not impossible to share OT data with the company’s own IT department, much less on the IIoT. The challenge is to fully protect the perimeter, and yet still provide access to the data from OPC UA servers inside.
Accessing OPC UA servers or any other industrial system from the IIoT should be done through a secure network connection. The typical approach, one that many take for granted, is to use a VPN (Virtual Private Network). VPN technology is well known, having been used for decades in the IT world. In essence, a VPN provides an outside user with a log-in to the network, and establishes a secure tunnel through the Internet to allow access to the system―the entire system. And that can lead to problems.
While OPC UA can work over a VPN, that doesn’t guarantee robust security. VPNs were not designed for use with industrial process control systems. In fact, they can open vulnerabilities even in the IT world. The attack on Target stores in North America that cost the company millions of dollars was perpetrated through a VPN. Hackers got hold of a user name and password, and gained access to the system. Once in, they quickly found their way to customer records and credit card numbers, and had a field day. The problem with using a VPN to access an industrial system is not only that every VPN user account is a potential access point, but that once someone is inside the perimeter they gain access to the whole system.
The drawbacks of using a VPN for the IoT are examined in detail by Clemens Vasters, a Microsoft Developer. In a paper titled Internet of Things: Is VPN a False Friend? Vasters said, “VPN provides a virtualized and private (isolated) network space. The secure tunnels are a mechanism to achieve an appropriately protected path into that space, but the space per-se is not secured, at all. It is indeed a feature that the established VPN space is fully transparent to all protocol and traffic above the link layer.”
Forward-thinking people who are working on the IIoT recognize this inherent risk in using VPNs. Many IT departments now require reverse proxies for OT systems to mask all internal servers and expose just one server to the Internet. But this approach does not secure OPC UA for the IIoT.
OPC UA clients can connect through reverse proxies using HTTP, but not HTTPS due to certificate handling. The proxy will either require opening a new firewall port, or effectively create a path to the control system that could easily be overlooked in the future. Either way an attack surface gets opened in the corporate perimeter. Furthermore, even if the message itself is encrypted, the message headers are exposed to outside observers. The only alternatives involve effectively tunneling through the proxy directly to the control system, which is what the proxy is trying to prevent.
The bottom line is that a reverse proxy is an improvement over a VPN, but it still requires a point of access into the control system from the Internet or IT network. Any point of access is an attack surface, and even if the server code is bulletproof it is still a candidate for a spear-phishing compromise.
The best way to completely close the plant perimeter is to eliminate all inbound connections, allowing only outbound connections. This is a good idea in principle, as it does not expose the plant to attack. The system presents zero attack surface, becoming invisible to hackers who cannot attack what they cannot see.
However, outbound connections run afoul of traditional design expectations. Effectively they turn the paradigm of industrial data communications on its head. Most client/server architectures, including OPC UA, assume that the server holds the data and the client initiates a connection to interact with it. The server is the authority on the data set, while the client is the non-authoritative user. Thus, in the OPC UA world-view the server must be situated with the primary data source, inside the control system.
To make a push design work in the IIoT, the server/client relationship must be reversed. The client must be the authority (inside the control system), and the server must be a non-authoritative receiver of the data. The client must be able to construct the data set on the server on the fly, based on its knowledge of the control system. This reversal of the client/server roles is something that OPC UA cannot accomplish on its own, but can be added through appropriate gateway software.
Using a push mechanism allows both OT and IT to completely close the network perimeter. If there is no way to make a connection from outside the network then there is no attack surface to exploit and there is no user to fool into revealing his password.
But even a closed perimeter is not sufficient. Best practice in IT networks is to route outgoing web traffic through a forward proxy, and to deny all other network traffic to the Internet. This substantially improves security by effectively shielding the internal network from a direct Internet connection. To be robust and IT-compliant the outbound IIoT connection must be able to pass through a standard forward proxy. Although OPC UA doesn’t inherently support forward proxies, appropriate gateway software can once again add this capability.
The Chatham House Report, Cyber Security at Civil Nuclear Facilities Understanding the Risks, points out an alarming lack of security at some of the most critical infrastructure installations in the world, and makes a number of design recommendations. At one point it states, “Many industrial control systems are insecure by design, since cyber security measures were not designed in from the beginning.” And this does not just apply to nuclear facilities. Indeed, the “many industrial systems” may well include those which now or soon might incorporate OPC UA. And they require a new approach, a new design for security on the IIoT.
The new design approach must allow OPC UA clients from any location to connect and acquire data from OPC UA servers within the plant perimeter, to eliminate the need for reverse proxies and VPNs and to avoid opening any inbound firewall ports. At the same time, to fully support OPC UA’s real-time data access, the design must support bi-directional data communication between OT and IT systems and across the Internet at speeds very close to network propagation times. Secure-by-design for the IIoT should take a no-compromise approach, offering the best possible combination of speed, security, and convenience.
With this level of security, and near-real-time speeds, there is one more design consideration: practicality. To gain traction among users, the design should be convenient to implement. It should, for example, allow for seamless integration with legacy installations using OPC Classic and other industrial protocols, as well as newer OPC UA-enabled systems. It should provide a loose coupling to the IIoT, one that allows remote, authorized and secure access the data, optionally including supervisory control, but that has no impact on the primary control system if it gets disconnected. And it should be easy enough to implement that it doesn’t overly tax the time or resources of the system integrator or plant engineer who is implementing it.
This is the kind of design that is needed to secure the IIoT, and make it compatible with today’s factory or process. OPC UA is the industrial protocol of the present, and of the future. It has the ability to integrate plant data from virtually any machine or device, large or small, as well as to bring the disparate worlds of OT and IT together. When OPC UA is wedded to the appropriate, secure-by-design IoT technology, it will play a key role in Industrie 4.0 and IIoT applications.