A platform engineered around Four Pillars of Dependability

Round trip latency within a datacenter:

< 30ms for 99th percentile

Transit latency measured at the Ably datacenter boundary. Measured from the point a message arrives from a publisher at an Ably datacenter to the point it leaves en-route to a subscriber within the same region.

Round trip latency from any of our 307 PoPs globally that receive at least 1% of our global traffic:

< 65ms for 99th percentile

This represents the transit latency individual clients experience. Measured at the Point of Presence (PoP) boundary within the Ably access network, which will be closer than a datacenter. Limited to PoPs that receive at least 1% of global traffic.

Mean roundtrip latency

< 99ms from all 307 PoPs

Transit latency for all PoPs, including those that are remote and rarely used, hence a 30ms higher latency target.

Channel (shard) throughput:

200 messages per second, 13MiB per second

Ably provides unlimited throughput capacity at a system-level. We achieve this with channels (sharding), constraining message and bandwidth throughput per second at a channel (shard) level in order to provide predictable performance. You can activate an unlimited number of channels for unlimited throughput.

Channel resource allocation

< 200ms for 99th percentile

Ably is a stateful system so there is a latency ‘cost’ when activating new channels. This latency has no material impact on the performance for subscribers and is different from message round-trip latency (< 65ms), which is the latency subscribers actually experience. It’s possible to activate channels ahead of time to bypass this initial resource allocation latency and increase predictability of latency for clients.

Channel churn rate

limitless (constrained only by quota)

This is the rate at which you can allocate and deallocate channels. This is effectively limitless: you can in theory activate one million channels per second.

Guaranteed Message Ordering from any single realtime or non-realtime publisher to all subscribers

Companies like HubSpot, Vitac, Genius Sports, and 17Media rely on Ably for message ordering to simplify app architecture and engineering. This also allows us to deliver features like Message Delta Compression while maintaining ordering of messages.

Idempotent publish operations are guaranteed within two minutes

We guarantee messages will be published only once as we discard those delivered multiple times. This provides flexibility around how you design your app as you don’t need to account for duplication. Limited to two minutes by default, but can reach back into persisted history for up to 72 hours.

Exactly-once delivery semantics with the Ably protocol

Ably’s exactly-once message delivery semantics mean you can simplify your app so it doesn’t need to account for duplicate or failed message deliveries, as is the case with at least once or at most once delivery. Note that this is dependent on using Ably client library SDKs that support idempotency in publishers and serial number based stream resumes in subscribers.

100% Guaranteed Message Delivery and Onwards Processing

Ably’s design and protocol ensures that once an ACK is received by the publishing client, all subscribers on that channel are guaranteed to receive the message.

Preserve connection integrity guarantee across disconnection for 2 minutes

Ably’s clients ensure connection state is maintained so abrupt disconnections or intermittent connections are resumed automatically by the SDKs, and message stream continuity ensured. Messages published when disconnected are delivered upon reconnection.

Maximum divergence of 30s advertised state and actual state of connection liveness & presence

The Ably protocol and underlying transports incorporate a liveness check that guarantees a faulty connection is detected within no more than 30s. Typically a dropped connection is detected immediately, however all transports depend on TCP/IP, which can in some circumstances fail to detect a connection failure. In these cases, the Ably liveness check ensures the unhealthy connection is determined, and all subscribers listening for connection or presence state will be notified within the 30s window.

Message survivability of 8x9s as a result of instance failures.

When Ably confirms receipt of a message (ACK), we guarantee we’ll process that. Once we receive a message, we immediately begin migrating it so we can replicate it in two Availability Zones to mitigate against instance failure. Failure of AZs is independent. If we drop down to one replica due to AZ failure we can be back to two replicas within 30 seconds.

Instance failure is calculated on the fact that any two instances failing within a five minute window of one another is 0.0000007%. And we’ve designed Ably so that any instance failure will migrate to a healthy instance within eight seconds, so we can have 99.999999% (8x9s) message availability.

Message survivability of 8x9s as a result of datacenter failure.

If there’s a problem causing issues within an AZ, for example a networking issue, we won’t be able to redistribute load within a datacenter. In this case, we fall back to datacenters in other AZs. We can survive two AZs going down simultaneously without bringing more AZs online.

Ably is designed around AZs with 99.99% SLAs, which statistically means we can provide 99.999999% (8x9s) message survivability.

Persisted data survivability of 10x9s as a result of instance or datacenter failures

This measures the reliability of our globally-available long-term storage. Once messages are persisted, we provide 99.99999999% (10x9s) survivability. You can continue to access data even if one or more regions globally might be down.

Edge network failure resolution by the client SDKs within 30s

Our SDKs can detect and resolve faults by finding a healthy datacenter within 30s.

Automated routing of all traffic away from an abrupt failure of datacenter in less than two minutes

We can detect and route away from abrupt failures in less than two minutes. Our routing layer will stop routing clients to that datacenter and route them elsewhere.

Discontinuity time for an emergency ATM response under two minutes

We have the ability to manually reroute traffic for customers with active traffic management within two mins. This is an enterprise feature.

50% global capacity margin for instantaneous surge

Ably operates at internet-scale, so our normal dimensions for capacity are already large. Regardless, we operate at 50% capacity margin so we can elastically deal with instant surges in demand and continue to be available in the event of AZ failure.

Connection capacity can double every 5 mins, halve every 10 mins & Channel capacity can double every 10 mins, halve every 20 mins

Ably can react to changes and elastically scale beyond instant surge capacity. But we must maintain state in all new areas when scaling. To do this and allow the system to keep up as it scales, we constrain the ability of the system to double (or halve) in capacity.

DoS: Layer 3, 4 and 7 defence in our edge network

Ably is designed with mechanisms to defend against various DoS vectors across different layers. This includes Layer 7 ‘attacks’ that might be legitimate operations but at unsustainable rates.

Max number of channels:

Limitless - constrained only by plan quota

Ably can scale limitlessly with an unlimited number of channels. We achieve this with channel sharding, where each channel has limited capacity, but you can activate as many channels as you need for your scale. For example, HubSpot employs over 500m channels per day on Ably.

Max number of connections:

Limitless - constrained only by quota

Ably can serve unlimited numbers of connections. This includes fanout to millions of subscribers over a handful of channels, or one-to-one connections for each user over individual channels.

Max message throughput:

Limitless - constrained only by quota

By restricting throughput on a per channel basis, we provide the ability to have unlimited throughput in aggregate. This is a mechanism to facilitate horizontal scaling.

Global service availability:

99.9999%

Ably is designed around the statistical probability that service availability will be 99.9999% (6x9s) - just 31s of downtime per year. To account for real-world behaviour, the lowest SLA we design around and commercially offer is 99.999% (5x9s) - 5 minutes 15 seconds of downtime per year.

Over the last year Ably’s uptime has been 100%, which is why we’re legitimately able to offer a 99.999% uptime SLA: status.ably.com.