Real architecture · Junior
Layered SLOs: how the BC measures PSP latency and availability
Splits end-to-end latency into segments (t0'…t6a), each with an owner and a percentile. Targets depend on PSP size: A 99.5%, B 99.0%, C 98.5%, D 95.0%. Failures of SPI, DICT, or RSFN do not count against the PSP.
Short answer
Splits end-to-end latency into segments (t0'…t6a), each with an owner and a percentile. Targets depend on PSP size: A 99.5%, B 99.0%, C 98.5%, D 95.0%. Failures of SPI, DICT, or RSFN do not count against the PSP. The Pix Timing Manual defines time and availability targets for the institutions participating in Pix.
01 · CONTEXT
Public targets for those who operate Pix
The Pix Timing Manual defines time and availability targets for the institutions participating in Pix. The manual calls these targets service level agreements (SLAs), and all indicators are calculated on a monthly basis.
Measuring well starts with good requisitos: what is measured, who measures it, and what is left out of the count.
There is also an absolute ceiling, which coexists with the percentile targets: a Pix sent to the SPI's primary message transmission channel has a maximum limit of 40 seconds, and a transaction that is not settled within that time is rejected.
02 · THE TIMELINE
Time milestones along the payment
The manual describes the settlement cycle of a Pix as a sequence of time milestones. Each event can be measured from both sides of an interaction, and the difference between the two measurements stems from the two agents' clocks and from the technology solution used in the communication.
For an individual, the payer's PSP receives the user's confirmation to send the payment order, before any security check and amount hold: this is milestone t0’.
The payer's PSP creates the pacs.008 message (milestone t1) and the SPI receives it through an HTTP request (milestone t1’).
The SPI makes the pacs.008 message available for consumption by the payee's PSP (milestone t2).
The payee's PSP replies with the pacs.002 message and the SPI receives it through an HTTP request (milestone t3’).
The SPI swaps the balances in the two PSPs' Instant Payment Accounts (milestone t4, the finality).
The SPI makes the pacs.002 message available for consumption by the payer's PSP (milestone t5a), and the payer's PSP sends the user the result notification (milestone t6a).
The finality milestone, where the SPI swaps the balances, appears on the timeline only for illustration: the manual says there is no SLA for this time.
03 · ONE INDICATOR PER SEGMENT
Each segment has an owner and a percentile
The manual separates the indicators of monitoramento by who computes them. The SPI calculates the time spent by the payer's PSP on Pix initiation and the time spent by the payee's PSP on authorization. The paying user's experience, measured from receipt of the order to notification of the result, is computed by the payer's PSP, and all participants, direct or indirect, report it.
| Indicador | Intervalo | Who measures | Percentiles and times |
|---|---|---|---|
| Pix initiation by the payer's PSP | from t0’ to t1 | SPI | 50%: 0.9 second; 95%: 1.5 seconds |
| Authorization by the payee's PSP | from t2 to t3’ | SPI | 50%: 1.4 seconds; 95%: 2.3 seconds |
| Paying user experience | from t0’ to t6a | Payer's PSP | 50%: 6.0 seconds; 99%: 10.0 seconds |
04 · WHO MEASURES, WITH WHAT CLOCK
The clock rule
The manual sets two premises for the measurement. Clocks of servers from different institutions cannot be used in the calculation of an indicator, because small differences could significantly distort the indicators. And, within a single institution, clocks must be synchronized to a level that ensures there will be no distortion; each institution is responsible for keeping its servers' clocks synchronized.
The SPI calculates the indicators for time spent by the payer's PSP on Pix initiation and by the payee's PSP on authorization.
The payer's PSP computes and reports the paying user's experience, measured between receipt of the order and notification of the result.
Clocks from different institutions cannot be used in the calculation of an indicator, because small differences could distort it significantly; each institution keeps its servers' clocks synchronized.
SLA indicators are calculated on a monthly basis.
05 · WHAT IS LEFT OUT OF THE COUNT
Exclusions of fault
As a general rule, all messages count, including error messages, but the indicators list their own exclusions. The paying user's experience does not count Scheduled Pix or operations whose settlement was interrupted by a payee PSP timeout in the SPI; the payee PSP's authorization indicator does not count operations interrupted by a timeout in the SPI.
The disponibilidade index follows a different logic: it is the ratio id = hf / hp between the hours of effective service operation for end users and the hours the service should be open, which are twenty-four per day, every day of the year. Only the transactional service counts, that is, transfers, payments, and receipts.
The index reflects the combined effect of all parts that are the participant's or its partners' responsibility, including the IT service provider and the settlement agent in the SPI, whatever the source of the unavailability.
Unavailability or instability of the SPI, the DICT, the RSFN, or another component that is the Central Bank's responsibility is not included in the calculation.
Connectivity failures of the end user, their device, or the apps installed on it are also left out.
Transactions rejected for lack of participant liquidity are also left out: only the technical aspects of the transactional service enter the index.
06 · TARGETS BY SIZE
The target depends on the participant's size
The monthly availability target varies with the participant's category. The category comes from its share of the total Pix transactions settled in the SPI in the last year, as payer or as payee, using the larger of the two sums. The participant stays in the category from July of the current year to June of the following year.
| Categoria | Share of total settled in the SPI | Monthly availability target |
|---|---|---|
| Categoria A | More than 2.0% | 99,5% |
| Categoria B | Between 2.0% and 0.5% | 99,0% |
| Categoria C | Between 0.5% and 0.1% | 98,5% |
| Categoria D | Other participants | 95,0% |
The indicators are combined into an overall index. Level 1 indicators, the most relevant to the Pix user experience, weigh more than level 2: the paying user's experience and availability are level 1, and the time spent by the payer's PSP on initiation is level 2. In the monthly index (IGAm), compliance counts level 1 with weight two and level 2 with weight one, and a recurrence deduction is subtracted from it. An indicator is recurrent if it was not met in the month of the calculation and in the immediately preceding month.
What serves compliance with the SLAs is the quarterly index (IGAt), the simple average of the monthly indices; the monthly one serves monitoring. A participant whose quarterly index equals or exceeds the reference value for its category is considered regular.
| Categoria | Meta | Reference value |
|---|---|---|
| Categoria A | 100% | 90% |
| Categoria B | 100% | 86% |
| Categoria C | 100% | 83% |
| Categoria D | 100% | 80% |
| Indirect participants | 100% | 65% |
07 · THE CENTRAL BANK SIDE
The targets of those who operate the infrastructure
The manual also includes service level agreements applied to the main BC systems that support Pix, calculated on a monthly basis. The SPI availability index has a target of 99.9%, considering a rolling window of the last three months. Unavailability is counted when 80% or more of settlement requests receive errors caused by the SPI or its infrastructure and this persists for more than 36 seconds. For key queries, the DICT has a time of 1 second at the 99th percentile.
Frequently asked questions
What is Layered SLOs: how the BC measures PSP latency and availability?
Splits end-to-end latency into segments (t0'…t6a), each with an owner and a percentile. Targets depend on PSP size: A 99.5%, B 99.0%, C 98.5%, D 95.0%. Failures of SPI, DICT, or RSFN do not count against the PSP.
What does the Layered SLOs: how the BC measures PSP latency and availability design cover?
It is organized in these sections: Public targets for those who operate Pix, Time milestones along the payment, Each segment has an owner and a percentile, The clock rule, Exclusions of fault, and The target depends on the participant's size.
How long is Layered SLOs: how the BC measures PSP latency and availability and what level is it?
It takes about 12 minutes to read and is rated Junior level.
Now design it yourself
Run this case as an interview in the System Design school: you make the decisions, the tutor challenges them and you see how your design scores.