System Design Interviews in Singapore: MAS Limits Applied
A timed system design method for Singapore, turning MAS's 4-hour downtime budget, 4-hour RTO and 1-hour incident notice into design choices.
By Redbison Dev · · 7 min read

Singapore edition. Written for Singapore interviews at banks, insurers, payment firms and their vendors: MAS's Notice on Technology Risk Management and its FAQ (critical systems, 4 hours of downtime in 12 months, RTO of 4 hours, 1-hour notice), the Singapore cloud region and its neighbours, and PDPA breach handling. US edition
In Singapore a system design round follows the usual four steps: scope and estimate, sketch the main path, go deep, then resilience and trade-offs. At a bank, insurer or payment service provider, the non-functional requirements arrive partly written by MAS. Its Notice on Technology Risk Management sets a downtime budget for each critical system, a maximum recovery time, and a 1-hour clock for telling MAS about a severe incident, and MAS's FAQ explains each with examples. This edition turns those into availability figures and design choices, then works through an online banking design. Practise with the prompts in the System Design Interview Framework (Singapore).
The method, timed
| Step | Minutes | Output |
|---|---|---|
| Requirements and estimates | 8 to 10 | Features, non-goals, throughput, storage, the MAS figures for critical systems |
| High-level design | 10 to 15 | Main path, APIs, data model |
| Detailed design | 15 to 20 | One or two components: schema, partitioning, failover, idempotency |
| Resilience and trade-offs | 8 to 10 | Downtime budget, detection, recovery, where replicas live |
The MAS figures, from the FAQ
MAS's FAQ on the Notice on Technology Risk Management gives the numbers and how to apply them:
- Critical systems. Examples include ATM systems, online banking systems, and systems that support payment, clearing or settlement functions (A5).
- Downtime budget. The maximum unscheduled downtime of each critical system must not exceed 4 hours within any period of 12 months (A10.2), and the count includes total, partial or intermittent disruption (A10.1).
- Recovery time. Systems identified as critical should have a recovery time objective of not more than 4 hours (A13).
- Notice. An institution must notify MAS within 1 hour of discovering a system malfunction or IT security incident with severe and widespread impact on its operations, or a material impact on customers, whenever it happens (A9).
- High availability counts. A failure recovered through a high-availability configuration that does not affect the proper functioning of the system need not be reported (A7).
The firm decides which systems are critical; the limits then apply to each one.
Turning 4 hours into an availability target
Four hours out of the 8,760 hours in a year is:
availability = 1 − 4 / 8,760 = 99.954%
But the window rolls. In MAS's own example, an outage of 3 hours on 1 July and one of 30 minutes on 20 December put the 12-month total at 3.5 hours; from the following 1 July the first outage drops out and the total is 0.5 hours until the December one expires. In interview terms, a single bad release can use most of a year's budget in an afternoon.
Detection eats into recovery. The FAQ's example (A14): if an incident starts at T and is discovered at T + 1, MAS must be told by T + 2, but the 4-hour recovery objective still runs from T, so the system must be back by T + 4. A design that takes an hour to notice a failure has only three hours to fix it.
Worked example: online banking for a Singapore bank
Prompt. Design the online banking backend: login, balances, transfers, transaction history.
Requirements. Treat it as a critical system. Non-functional: 4 hours or less of unscheduled downtime in any 12 months, partial outages included; RTO of 4 hours or less; failures detected and triaged within minutes so the 1-hour notice is met when needed.
Estimates (premises to agree):
logged-in peak: 20,000 requests a second (pay day, assumed)
transfers: 5% of requests → 1,000 writes a second at peak
history: ~500 bytes per transaction, 2 years kept online
daily transfers: 3 million a day (premise; the 1,000/s peak is not the daily rate)
storage: 3,000,000 x 500 B x 730 days ≈ 1.1 TB online, before replicas
budget: 4 h / 12 months; already used this year: 1.5 h (premise)
remaining budget: 2.5 hours
Regions. The AWS regions list shows Asia Pacific (Singapore) ap-southeast-1 with three availability zones, and Asia Pacific (Malaysia) ap-southeast-5 and Asia Pacific (Jakarta) ap-southeast-3 as nearby opt-in regions, each also with three zones.
Design choices that follow.
- Spread across the three Singapore zones with synchronous replication for the transfer ledger. A zone failure absorbed this way is the kind of high-availability recovery that, per the FAQ, does not need reporting, and it does not spend the downtime budget if customers see no disruption.
- Make partial failure visible. Because partial and intermittent disruption counts, monitor customer-facing success rates per journey (login, balance, transfer), not just host health. A transfer path failing 30% of requests is downtime.
- Release safely. With 2.5 hours left in the budget, canary releases, feature flags and one-step rollback are not polish, they are how you stay inside the limit.
- Cross-border recovery is a requirement question. A copy in Malaysia or Jakarta gives regional separation but moves data outside Singapore. Ask whether that is allowed, and if it is, replicate asynchronously and state the data loss on failover.
- Idempotent transfers. Retries after a timeout must not move money twice; give each transfer a client-generated ID and enforce uniqueness in the ledger.
Personal data
Online banking holds personal data, so the PDPA's data breach obligations apply alongside MAS's. The PDPC's Guide on Managing and Notifying Data Breaches under the PDPA (revised 15 March 2021) sets two clocks. Once there are credible grounds to believe a breach has occurred, the organisation must take reasonable and expeditious steps to assess whether it is notifiable within 30 calendar days. If it is, the PDPC must be told as soon as practicable and no later than 3 calendar days after that determination. A breach is notifiable if it is likely to cause significant harm or involves the personal data of 500 or more individuals; where the number is unknown, the guide says to notify once there is reason to believe it is at least 500.
That 500 threshold is a design requirement. Audit logs queryable by customer and date, and a data map from each store to the customers in it, let you count affected customers within hours rather than spend the 30-day window on it.
Common mistakes in Singapore rounds
- Stating "99.95% availability" without saying that the MAS budget rolls over 12 months and includes partial disruption.
- Treating the 1-hour notice and the 4-hour RTO as the same clock.
- Proposing a disaster recovery region outside Singapore without asking whether data may leave.
- Ignoring detection time; slow alerting turns a 4-hour objective into a 3-hour one.
Tools for this job
- System Design Interview Framework (Singapore): the four steps, practice prompts and topic checklist.
- Diagram Tool: draw the design and redraw it from memory.
- Coding Interview Prep Checklist: for coding rounds in the same loop.
- STAR Answer Builder: for the behavioural round, with the STAR method guide.
Frequently asked questions
What is the MAS downtime limit for critical systems?
The maximum unscheduled downtime of each critical system must not exceed 4 hours within any 12-month period, counting total, partial and intermittent disruption, according to MAS's FAQ on the Notice on Technology Risk Management. That is about 99.954% availability over a rolling year.
How is the 1-hour notice different from the 4-hour RTO?
The RTO runs from the start of the disruption; the 1-hour notice runs from discovery. If an incident starts at T and is found at T + 1, MAS must be told by T + 2 and the system recovered by T + 4.
Does a failover need to be reported to MAS?
Not if the failure was recovered through a high-availability configuration and the system kept functioning properly, according to the FAQ. Reportable events are those with severe and widespread impact on operations or a material impact on customers.
Which systems count as critical?
MAS gives ATM systems, online banking systems and systems supporting payment, clearing or settlement as examples. Each institution identifies its own critical systems against the Notice's criteria.
Sources and further reading
- FAQ: Notice on Technology Risk Management, Monetary Authority of Singapore. Critical systems, the 4-hour downtime budget and its rolling example, the 4-hour RTO, the 1-hour notice and high availability.
- Guidelines on Risk Management Practices: Technology Risk, Monetary Authority of Singapore. The wider technology risk guidelines, published 18 January 2021.
- Guide on Managing and Notifying Data Breaches under the PDPA, Personal Data Protection Commission, revised 15 March 2021. The 30-day assessment, the 3-day notice and the 500-person threshold.
- AWS Regions, Amazon Web Services. Singapore ap-southeast-1, Malaysia ap-southeast-5 and Jakarta ap-southeast-3, three availability zones each.
- Brewer's conjecture and the feasibility of consistent, available, partition-tolerant web services, ACM SIGACT News. Gilbert and Lynch's proof of the CAP theorem.
Open the System Design Interview Framework
More Interview Prep
Every tool in the Interview Prep section, free and in your browser.