Skip to content
Browse the docs

Assumptions

View on GitHubReport a problem with this page

Every number ArchSim 1.1.1 uses without you typing it, and where it comes from:

  • documented: the vendor's documentation states it (linked);
  • illustrative: a typical value ArchSim chose, not a vendor guarantee and not a measurement. Replace it with your own numbers before trusting absolute results.

Generated from the code that uses these values (./gradlew :archsim-engine:generate); a test fails when this file is out of date.

Technology tiers#

Core technologies are the ones ArchSim focuses on and will validate against measurements first. Experimental ones stay available, with a badge in the parts library, the inspector, reports and exports that says: "Behaviour based on documentation and illustrative numbers; not yet validated against measurements." The generic part of each kind is core: it adds no technology behaviour, only the shared mechanisms (workers, connections, queues, service times) that the core technologies use too. No technology has been validated against measurements yet.

Part Core Experimental
Load Balancer nginx, aws_alb, generic envoy, api_gateway, kong, haproxy, traefik
Service / API Servers nodejs, java, go, generic dotnet, rust, python, php, ruby
Cache redis, generic valkey, memcached, cdn, varnish
Database postgresql, generic mysql, mariadb, oracle, sqlserver, cockroachdb, yugabytedb, mongodb, couchbase, cassandra, scylladb, dynamodb, elasticsearch, opensearch
Message Queue / Stream kafka, sqs, generic rabbitmq, pubsub, nats, redis_streams
Kubernetes cluster kubernetes eks, gke, aks, k3s, openshift

Technology behaviour defaults#

What each technology sets for its "auto" settings (Technology behaviour in the inspector). Setting one on a part overrides it.

Load Balancer#

Technology Setting Value Basis
nginx, core Upstream timeout (upstreamTimeoutMs) 60,000 ms documented: nginx.org
envoy, experimental Upstream timeout (upstreamTimeoutMs) 15,000 ms documented: envoyproxy.io
Circuit breaker: max requests per target (circuitBreakerMaxRequests) 1,024 documented: envoyproxy.io
aws_alb, core Upstream timeout (upstreamTimeoutMs) 60,000 ms documented: aws.amazon.com
api_gateway, experimental Upstream timeout (upstreamTimeoutMs) 29,000 ms documented: aws.amazon.com
kong, experimental Upstream timeout (upstreamTimeoutMs) 60,000 ms documented: developer.konghq.com
haproxy, experimental Upstream timeout (upstreamTimeoutMs) none illustrative
traefik, experimental Upstream timeout (upstreamTimeoutMs) none illustrative
generic, core Upstream timeout (upstreamTimeoutMs) none illustrative

Service / API Servers#

Technology Setting Value Basis
nodejs, core Concurrency model (runtimeModel) event_loop documented: nodejs.org
GC pause (gcPauseMs) 2 ms illustrative
Requests between GCs (gcEveryRequests) 1,500 illustrative
JIT warm-up slowdown (warmupFactor) 1 illustrative
JIT warm-up time (warmupS) 0 s illustrative
Cold start (coldStartMs) lognormal, median 250 ms, p95 700 ms illustrative
java, core Concurrency model (runtimeModel) thread_pool documented: spring.io
GC pause (gcPauseMs) 12 ms illustrative
Requests between GCs (gcEveryRequests) 2,000 illustrative
JIT warm-up slowdown (warmupFactor) 2.5 illustrative
JIT warm-up time (warmupS) 45 s illustrative
Cold start (coldStartMs) lognormal, median 1,200 ms, p95 3,500 ms illustrative
dotnet, experimental Concurrency model (runtimeModel) thread_pool illustrative
GC pause (gcPauseMs) 6 ms illustrative
Requests between GCs (gcEveryRequests) 2,000 illustrative
JIT warm-up slowdown (warmupFactor) 1.5 illustrative
JIT warm-up time (warmupS) 20 s illustrative
Cold start (coldStartMs) lognormal, median 600 ms, p95 1,500 ms illustrative
go, core Concurrency model (runtimeModel) thread_pool documented: go.dev
GC pause (gcPauseMs) 0.1 ms documented: go.dev
Requests between GCs (gcEveryRequests) 1,000 illustrative
JIT warm-up slowdown (warmupFactor) 1 illustrative
JIT warm-up time (warmupS) 0 s illustrative
Cold start (coldStartMs) lognormal, median 100 ms, p95 300 ms illustrative
rust, experimental Concurrency model (runtimeModel) thread_pool illustrative
GC pause (gcPauseMs) 0 ms documented: doc.rust-lang.org
JIT warm-up slowdown (warmupFactor) 1 documented: doc.rust-lang.org
JIT warm-up time (warmupS) 0 s documented: doc.rust-lang.org
Cold start (coldStartMs) lognormal, median 60 ms, p95 200 ms illustrative
python, experimental Concurrency model (runtimeModel) thread_pool documented: gunicorn.org
GC pause (gcPauseMs) 1 ms illustrative
Requests between GCs (gcEveryRequests) 2,000 illustrative
JIT warm-up slowdown (warmupFactor) 1 illustrative
JIT warm-up time (warmupS) 0 s illustrative
Cold start (coldStartMs) lognormal, median 250 ms, p95 700 ms illustrative
php, experimental Concurrency model (runtimeModel) thread_pool documented: php.net
GC pause (gcPauseMs) 0 ms illustrative
JIT warm-up slowdown (warmupFactor) 1 illustrative
JIT warm-up time (warmupS) 0 s illustrative
Cold start (coldStartMs) lognormal, median 300 ms, p95 900 ms illustrative
ruby, experimental Concurrency model (runtimeModel) thread_pool illustrative
GC pause (gcPauseMs) 3 ms illustrative
Requests between GCs (gcEveryRequests) 1,500 illustrative
JIT warm-up slowdown (warmupFactor) 1 illustrative
JIT warm-up time (warmupS) 0 s illustrative
Cold start (coldStartMs) lognormal, median 300 ms, p95 900 ms illustrative
generic, core Concurrency model (runtimeModel) thread_pool illustrative
GC pause (gcPauseMs) 0 ms illustrative
JIT warm-up slowdown (warmupFactor) 1 illustrative
JIT warm-up time (warmupS) 0 s illustrative
Cold start (coldStartMs) lognormal, median 300 ms, p95 900 ms illustrative

Cache#

Technology Setting Value Basis
redis, core Single-threaded commands (singleThreaded) on documented: redis.io
Eviction policy (evictionPolicy) allkeys_lru illustrative
Persistence (persistence) none illustrative
valkey, experimental Single-threaded commands (singleThreaded) on documented: valkey.io
Eviction policy (evictionPolicy) allkeys_lru illustrative
Persistence (persistence) none illustrative
memcached, experimental Single-threaded commands (singleThreaded) off documented: memcached.org
Eviction policy (evictionPolicy) allkeys_lru documented: memcached.org
Persistence (persistence) none documented: memcached.org
cdn, experimental Single-threaded commands (singleThreaded) off illustrative
Eviction policy (evictionPolicy) allkeys_lru illustrative
varnish, experimental Single-threaded commands (singleThreaded) off illustrative
Eviction policy (evictionPolicy) allkeys_lru illustrative
generic, core Single-threaded commands (singleThreaded) off illustrative
Eviction policy (evictionPolicy) allkeys_lru illustrative

Database#

Technology Setting Value Basis
postgresql, core Row locks on writes (rowLocks) on documented: postgresql.org
Reads wait for writers (readsWaitForWrites) off documented: postgresql.org
Commits wait for the log flush (commitFlush) on documented: postgresql.org
Replica writes (replication) async documented: postgresql.org
Replica apply workers (replicaApplyWorkers) 1 documented: postgresql.org
Reads served by (readFrom) all illustrative
mysql, experimental Row locks on writes (rowLocks) on illustrative
Reads wait for writers (readsWaitForWrites) off illustrative
Commits wait for the log flush (commitFlush) on illustrative
Replica writes (replication) async illustrative
Replica apply workers (replicaApplyWorkers) 4 documented: dev.mysql.com
Reads served by (readFrom) all illustrative
mariadb, experimental Row locks on writes (rowLocks) on illustrative
Reads wait for writers (readsWaitForWrites) off illustrative
Commits wait for the log flush (commitFlush) on illustrative
Replica writes (replication) async illustrative
Replica apply workers (replicaApplyWorkers) 4 illustrative
Reads served by (readFrom) all illustrative
oracle, experimental Row locks on writes (rowLocks) on illustrative
Reads wait for writers (readsWaitForWrites) off illustrative
Commits wait for the log flush (commitFlush) on illustrative
Replica writes (replication) async illustrative
Replica apply workers (replicaApplyWorkers) 4 illustrative
Reads served by (readFrom) all illustrative
sqlserver, experimental Row locks on writes (rowLocks) on illustrative
Reads wait for writers (readsWaitForWrites) on documented: learn.microsoft.com
Commits wait for the log flush (commitFlush) on illustrative
Replica writes (replication) async illustrative
Replica apply workers (replicaApplyWorkers) 4 illustrative
Reads served by (readFrom) all illustrative
cockroachdb, experimental Row locks on writes (rowLocks) on illustrative
Reads wait for writers (readsWaitForWrites) off illustrative
Commits wait for the log flush (commitFlush) on illustrative
Replica writes (replication) async illustrative
Replica apply workers (replicaApplyWorkers) 4 illustrative
Reads served by (readFrom) all illustrative
yugabytedb, experimental Row locks on writes (rowLocks) on illustrative
Reads wait for writers (readsWaitForWrites) off illustrative
Commits wait for the log flush (commitFlush) on illustrative
Replica writes (replication) async illustrative
Replica apply workers (replicaApplyWorkers) 4 illustrative
Reads served by (readFrom) all illustrative
mongodb, experimental Row locks on writes (rowLocks) on documented: mongodb.com
Reads wait for writers (readsWaitForWrites) off illustrative
Commits wait for the log flush (commitFlush) on illustrative
Replica writes (replication) majority documented: mongodb.com
Replica apply workers (replicaApplyWorkers) 16 illustrative
Reads served by (readFrom) primary documented: mongodb.com
Concurrent reads (tickets) (readTickets) 128 documented: mongodb.com
Concurrent writes (tickets) (writeTickets) 128 documented: mongodb.com
Router / coordinator time (routerOverheadMs) 0.5 ms illustrative
couchbase, experimental Row locks on writes (rowLocks) off illustrative
Commits wait for the log flush (commitFlush) off illustrative
Replica writes (replication) async illustrative
Replica apply workers (replicaApplyWorkers) 4 illustrative
Reads served by (readFrom) primary illustrative
cassandra, experimental Row locks on writes (rowLocks) off documented: cassandra.apache.org
Commits wait for the log flush (commitFlush) off documented: cassandra.apache.org
Replica writes (replication) none illustrative
Reads served by (readFrom) all illustrative
SSTables per cold read (sstablesPerRead) 2 illustrative
scylladb, experimental Row locks on writes (rowLocks) off illustrative
Commits wait for the log flush (commitFlush) off illustrative
Replica writes (replication) none illustrative
Reads served by (readFrom) all illustrative
SSTables per cold read (sstablesPerRead) 1.5 illustrative
dynamodb, experimental Row locks on writes (rowLocks) off illustrative
Commits wait for the log flush (commitFlush) off illustrative
Replica writes (replication) none illustrative
Reads served by (readFrom) all illustrative
Capacity units and partition limits (capacityUnits) on documented: aws.amazon.com
elasticsearch, experimental Row locks on writes (rowLocks) off illustrative
Commits wait for the log flush (commitFlush) off illustrative
Replica writes (replication) all documented: elastic.co
Replica apply workers (replicaApplyWorkers) 4 illustrative
Reads served by (readFrom) all illustrative
Reads sent to every shard (scatterReads) 1 documented: elastic.co
coordinatorMs 1 illustrative
opensearch, experimental Row locks on writes (rowLocks) off illustrative
Commits wait for the log flush (commitFlush) off illustrative
Replica writes (replication) all documented: elastic.co
Replica apply workers (replicaApplyWorkers) 4 illustrative
Reads served by (readFrom) all illustrative
Reads sent to every shard (scatterReads) 1 documented: elastic.co
coordinatorMs 1 illustrative
generic, core Row locks on writes (rowLocks) off illustrative
Reads wait for writers (readsWaitForWrites) off illustrative
Commits wait for the log flush (commitFlush) off illustrative
Replica writes (replication) none illustrative
Reads served by (readFrom) all illustrative

Message Queue / Stream#

Technology Setting Value Basis
kafka, core Partitioning (partitioning) key documented: kafka.apache.org
Publish acknowledgement (acks) all documented: kafka.apache.org
Batching linger (lingerMs) 5 ms documented: kafka.apache.org
Delivery guarantee (delivery) at_least_once illustrative
Visibility timeout (visibilityTimeoutS) none illustrative
Memory alarm blocks publishers (flowControl) off illustrative
rabbitmq, experimental Partitioning (partitioning) none illustrative
Publish acknowledgement (acks) leader illustrative
Batching linger (lingerMs) 0 ms illustrative
Delivery guarantee (delivery) at_least_once illustrative
Visibility timeout (visibilityTimeoutS) none illustrative
Memory alarm blocks publishers (flowControl) on documented: rabbitmq.com
sqs, core Partitioning (partitioning) none illustrative
Publish acknowledgement (acks) leader illustrative
Batching linger (lingerMs) 0 ms illustrative
Delivery guarantee (delivery) at_least_once illustrative
Visibility timeout (visibilityTimeoutS) 30 s documented: aws.amazon.com
Memory alarm blocks publishers (flowControl) off illustrative
pubsub, experimental Partitioning (partitioning) none illustrative
Publish acknowledgement (acks) leader illustrative
Batching linger (lingerMs) 0 ms illustrative
Delivery guarantee (delivery) at_least_once illustrative
Visibility timeout (visibilityTimeoutS) 10 s documented: cloud.google.com
Memory alarm blocks publishers (flowControl) off illustrative
nats, experimental Partitioning (partitioning) none illustrative
Publish acknowledgement (acks) leader illustrative
Batching linger (lingerMs) 0 ms illustrative
Delivery guarantee (delivery) at_least_once illustrative
Visibility timeout (visibilityTimeoutS) none illustrative
Memory alarm blocks publishers (flowControl) off illustrative
redis_streams, experimental Partitioning (partitioning) none illustrative
Publish acknowledgement (acks) leader illustrative
Batching linger (lingerMs) 0 ms illustrative
Memory alarm blocks publishers (flowControl) off illustrative
generic, core Partitioning (partitioning) none illustrative
Publish acknowledgement (acks) leader illustrative
Batching linger (lingerMs) 0 ms illustrative
Memory alarm blocks publishers (flowControl) off illustrative

Preset numbers#

The numbers a technology preset fills in when you add it from the parts library. Prices are illustrative until real cloud prices are added.

Load Balancer#

Preset Setting Value Basis
Nginx, core CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 2 GiB illustrative
Balancing policy (policy) round_robin illustrative
Max connections (maxConnections) 1,024 illustrative
Overhead (overheadMs) 0.2 ms illustrative
HAProxy, experimental CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 2 GiB illustrative
Balancing policy (policy) least_connections illustrative
Overhead (overheadMs) 0.15 ms illustrative
Envoy, experimental CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 2 GiB illustrative
Balancing policy (policy) least_connections illustrative
Overhead (overheadMs) 0.3 ms illustrative
Kong API gateway, experimental CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 4 GiB illustrative
Overhead (overheadMs) lognormal, median 1 ms, p95 3 ms illustrative
Rate limit (rateLimitRps) 1,000 req/s illustrative
Burst (rateLimitBurst) 200 req illustrative
Cost per hour (costPerHour) 0.05 $/h illustrative
ALB-style load balancer, core Overhead (overheadMs) lognormal, median 1.5 ms, p95 5 ms illustrative
Max connections (maxConnections) 100,000 illustrative
Cost per hour (costPerHour) 0.0225 $/h illustrative
API-gateway-style (managed), experimental Overhead (overheadMs) lognormal, median 6 ms, p95 20 ms illustrative
Rate limit (rateLimitRps) 10,000 req/s documented: aws.amazon.com
Burst (rateLimitBurst) 5,000 req documented: aws.amazon.com
Cost per hour (costPerHour) 0 $/h illustrative

Service / API Servers#

Preset Setting Value Basis
Generic service (VM), core CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 4 GiB illustrative
Platform (platform) vm illustrative
Java/Spring on Docker, core CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 4 GiB illustrative
Memory per request (memoryPerRequestMb) 1 MiB illustrative
Memory at idle (memoryBaselineMb) 768 MiB illustrative
Platform (platform) docker illustrative
Replicas (replicas) 2 illustrative
Workers per replica (workersPerReplica) 200 documented: spring.io
Queue limit per replica (queueCapacity) 100 documented: spring.io
Service time (serviceTimeMs) lognormal, median 12 ms, p95 40 ms illustrative
Workers block on downstream calls (blockingDownstream) yes illustrative
Cost per replica-hour (costPerHour) 0.17 $/h illustrative
Node.js on Docker, core CPU (cpuCores) 1 vCPU illustrative
Memory (memoryGib) 1 GiB illustrative
Memory per request (memoryPerRequestMb) 0.5 MiB illustrative
Memory at idle (memoryBaselineMb) 128 MiB illustrative
Platform (platform) docker illustrative
Replicas (replicas) 4 illustrative
Workers per replica (workersPerReplica) 1 documented: nodejs.org
Queue limit per replica (queueCapacity) 1,000 illustrative
Service time (serviceTimeMs) lognormal, median 4 ms, p95 12 ms illustrative
Workers block on downstream calls (blockingDownstream) no illustrative
Cost per replica-hour (costPerHour) 0.05 $/h illustrative
Go service on Kubernetes (HPA), core CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 1 GiB illustrative
Memory per request (memoryPerRequestMb) 0.25 MiB illustrative
Memory at idle (memoryBaselineMb) 64 MiB illustrative
Platform (platform) kubernetes illustrative
Replicas (replicas) 2 illustrative
Workers per replica (workersPerReplica) 64 illustrative
Queue limit per replica (queueCapacity) 200 illustrative
Service time (serviceTimeMs) lognormal, median 5 ms, p95 15 ms illustrative
Autoscaling (autoscaling) yes illustrative
Min replicas (minReplicas) 2 illustrative
Max replicas (maxReplicas) 10 illustrative
Target utilization (targetUtilization) 0.6 illustrative
Evaluation interval (scaleIntervalS) 15 s documented: kubernetes.io
Replica start-up time (startupS) 10 s illustrative
Cost per replica-hour (costPerHour) 0.05 $/h illustrative
Python service on Kubernetes (HPA), experimental CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 2 GiB illustrative
Memory per request (memoryPerRequestMb) 4 MiB illustrative
Memory at idle (memoryBaselineMb) 512 MiB illustrative
Platform (platform) kubernetes illustrative
Replicas (replicas) 2 illustrative
Workers per replica (workersPerReplica) 4 illustrative
Queue limit per replica (queueCapacity) 64 illustrative
Service time (serviceTimeMs) lognormal, median 15 ms, p95 60 ms illustrative
Autoscaling (autoscaling) yes illustrative
Min replicas (minReplicas) 2 illustrative
Max replicas (maxReplicas) 12 illustrative
Target utilization (targetUtilization) 0.6 illustrative
Evaluation interval (scaleIntervalS) 15 s documented: kubernetes.io
Replica start-up time (startupS) 25 s illustrative
Cost per replica-hour (costPerHour) 0.05 $/h illustrative
Serverless function (Lambda-style), core CPU (cpuCores) 1 vCPU documented: aws.amazon.com
Memory (memoryGib) 1.75 GiB documented: aws.amazon.com
Platform (platform) serverless illustrative
Concurrency limit (concurrencyLimit) 1,000 documented: aws.amazon.com
Keep warm (keepWarmS) 300 s illustrative
Service time (serviceTimeMs) lognormal, median 20 ms, p95 60 ms illustrative
Cost per replica-hour (costPerHour) 0.06 $/h illustrative

Cache#

Preset Setting Value Basis
Redis, core CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 6.4 GiB illustrative
Command executors (concurrency) 1 documented: redis.io
Cost per hour (costPerHour) 0.1 $/h illustrative
Valkey, experimental CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 6.4 GiB illustrative
Command executors (concurrency) 1 documented: valkey.io
Command time (serviceTimeMs) exponential, mean 0.035 ms illustrative
Cost per hour (costPerHour) 0.1 $/h illustrative
Memcached, experimental CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 6.4 GiB illustrative
Command executors (concurrency) 4 documented: memcached.org
Command time (serviceTimeMs) exponential, mean 0.04 ms illustrative
Cost per hour (costPerHour) 0.09 $/h illustrative
CDN edge cache, experimental Command executors (concurrency) 1,024 illustrative
Queue limit (queueCapacity) 100,000 illustrative
Hit ratio (hitRatio) 0.9 illustrative
Command time (serviceTimeMs) lognormal, median 1.5 ms, p95 4 ms illustrative
Cost per hour (costPerHour) 0.2 $/h illustrative
Varnish, experimental CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 8 GiB illustrative
Command executors (concurrency) 256 illustrative
Hit ratio (hitRatio) 0.85 illustrative
Command time (serviceTimeMs) lognormal, median 0.3 ms, p95 1.2 ms illustrative
Cost per hour (costPerHour) 0.08 $/h illustrative

Database#

Preset Setting Value Basis
PostgreSQL, core CPU (cpuCores) 4 vCPU illustrative
Memory (memoryGib) 16 GiB illustrative
Storage (storageGib) 200 GiB illustrative
Max connections per instance (maxConnections) 20 illustrative
Read query time (readServiceTimeMs) lognormal, median 4 ms, p95 15 ms illustrative
Write query time (writeServiceTimeMs) lognormal, median 7 ms, p95 28 ms illustrative
Cost per instance-hour (costPerHour) 0.35 $/h illustrative
PostgreSQL + 2 read replicas, core CPU (cpuCores) 4 vCPU illustrative
Memory (memoryGib) 16 GiB illustrative
Storage (storageGib) 200 GiB illustrative
Max connections per instance (maxConnections) 20 illustrative
Read replicas (readReplicas) 2 illustrative
Read query time (readServiceTimeMs) lognormal, median 4 ms, p95 15 ms illustrative
Write query time (writeServiceTimeMs) lognormal, median 7 ms, p95 28 ms illustrative
Cost per instance-hour (costPerHour) 0.35 $/h illustrative
MySQL, experimental CPU (cpuCores) 4 vCPU illustrative
Memory (memoryGib) 16 GiB illustrative
Storage (storageGib) 200 GiB illustrative
Max connections per instance (maxConnections) 20 illustrative
Read query time (readServiceTimeMs) lognormal, median 4 ms, p95 16 ms illustrative
Write query time (writeServiceTimeMs) lognormal, median 8 ms, p95 32 ms illustrative
Cost per instance-hour (costPerHour) 0.33 $/h illustrative
SQL Server, experimental CPU (cpuCores) 4 vCPU illustrative
Memory (memoryGib) 32 GiB illustrative
Storage (storageGib) 250 GiB illustrative
Max connections per instance (maxConnections) 30 illustrative
Read query time (readServiceTimeMs) lognormal, median 5 ms, p95 20 ms illustrative
Write query time (writeServiceTimeMs) lognormal, median 9 ms, p95 36 ms illustrative
Cost per instance-hour (costPerHour) 0.9 $/h illustrative
CockroachDB-style distributed SQL, experimental CPU (cpuCores) 4 vCPU illustrative
Memory (memoryGib) 16 GiB illustrative
Storage (storageGib) 300 GiB illustrative
Max connections per instance (maxConnections) 20 illustrative
Shards (partitions) (shards) 3 illustrative
Replication factor (replicationFactor) 3 documented: cockroachlabs.com
Consistency level (consistency) quorum illustrative
Read query time (readServiceTimeMs) lognormal, median 6 ms, p95 25 ms illustrative
Write query time (writeServiceTimeMs) lognormal, median 13 ms, p95 55 ms illustrative
Cost per instance-hour (costPerHour) 0.5 $/h illustrative
MongoDB replica set, experimental CPU (cpuCores) 4 vCPU illustrative
Memory (memoryGib) 16 GiB illustrative
Storage (storageGib) 250 GiB illustrative
Max connections per instance (maxConnections) 500 illustrative
Wait queue limit (queueCapacity) 1,000 illustrative
Read replicas (readReplicas) 2 illustrative
Reads served by (readFrom) replicas illustrative
Read query time (readServiceTimeMs) lognormal, median 3 ms, p95 12 ms illustrative
Write query time (writeServiceTimeMs) lognormal, median 5 ms, p95 20 ms illustrative
Cost per instance-hour (costPerHour) 0.3 $/h illustrative
MongoDB sharded cluster, experimental CPU (cpuCores) 4 vCPU illustrative
Memory (memoryGib) 16 GiB illustrative
Storage (storageGib) 250 GiB illustrative
Max connections per instance (maxConnections) 500 illustrative
Wait queue limit (queueCapacity) 1,000 illustrative
Shards (partitions) (shards) 4 illustrative
Read replicas (readReplicas) 2 illustrative
Reads sent to every shard (scatterReads) 0.1 illustrative
Read query time (readServiceTimeMs) lognormal, median 3 ms, p95 12 ms illustrative
Write query time (writeServiceTimeMs) lognormal, median 5 ms, p95 20 ms illustrative
Cost per instance-hour (costPerHour) 0.3 $/h illustrative
Cassandra, experimental CPU (cpuCores) 8 vCPU illustrative
Memory (memoryGib) 32 GiB illustrative
Storage (storageGib) 1,000 GiB illustrative
Max connections per instance (maxConnections) 32 illustrative
Shards (partitions) (shards) 6 illustrative
Replication factor (replicationFactor) 3 illustrative
Consistency level (consistency) quorum illustrative
Read query time (readServiceTimeMs) lognormal, median 4 ms, p95 18 ms illustrative
Write query time (writeServiceTimeMs) lognormal, median 1.5 ms, p95 6 ms illustrative
Cost per instance-hour (costPerHour) 0.4 $/h illustrative
DynamoDB-style key-value store, experimental Max connections per instance (maxConnections) 5,000 illustrative
Wait queue limit (queueCapacity) 1,000 illustrative
Capacity mode (capacityMode) on_demand illustrative
Read query time (readServiceTimeMs) lognormal, median 3 ms, p95 8 ms illustrative
Write query time (writeServiceTimeMs) lognormal, median 5 ms, p95 12 ms illustrative
Cost per instance-hour (costPerHour) 0.5 $/h illustrative
Elasticsearch, experimental CPU (cpuCores) 4 vCPU illustrative
Memory (memoryGib) 32 GiB illustrative
Storage (storageGib) 500 GiB illustrative
Max connections per instance (maxConnections) 7 documented: elastic.co
Wait queue limit (queueCapacity) 1,000 documented: elastic.co
Shards (partitions) (shards) 3 illustrative
Read replicas (readReplicas) 1 illustrative
Read query time (readServiceTimeMs) lognormal, median 12 ms, p95 50 ms illustrative
Write query time (writeServiceTimeMs) lognormal, median 20 ms, p95 100 ms illustrative
Cost per instance-hour (costPerHour) 0.45 $/h illustrative

External API#

Preset Setting Value Basis
Payments API Response time (latencyMs) lognormal, median 250 ms, p95 900 ms illustrative
Error rate (errorRate) 0.002 illustrative
Rate limit (rateLimitRps) 100 req/s illustrative
Cost per million calls (costPerMillionRequests) 0 $ illustrative
Email API Response time (latencyMs) lognormal, median 120 ms, p95 400 ms illustrative
Error rate (errorRate) 0.001 illustrative
Rate limit (rateLimitRps) 50 req/s illustrative
Cost per million calls (costPerMillionRequests) 1,000 $ illustrative
SMS API Response time (latencyMs) lognormal, median 300 ms, p95 1,200 ms illustrative
Error rate (errorRate) 0.005 illustrative
Rate limit (rateLimitRps) 30 req/s illustrative
Cost per million calls (costPerMillionRequests) 7,500 $ illustrative
Identity provider Response time (latencyMs) lognormal, median 80 ms, p95 250 ms illustrative
Error rate (errorRate) 0.0005 illustrative
Rate limit (rateLimitRps) 200 req/s illustrative
Cost per million calls (costPerMillionRequests) 0 $ illustrative
LLM inference API Response time (latencyMs) lognormal, median 1,500 ms, p95 6,000 ms illustrative
Error rate (errorRate) 0.01 illustrative
Rate limit (rateLimitRps) 50 req/s illustrative
Concurrency limit (concurrencyLimit) 100 illustrative
Cost per million calls (costPerMillionRequests) 2,000 $ illustrative
Object storage (S3-compatible) Response time (latencyMs) lognormal, median 25 ms, p95 80 ms illustrative
Error rate (errorRate) 0.0001 illustrative
Rate limit (rateLimitRps) 5,500 req/s documented: aws.amazon.com
Cost per million calls (costPerMillionRequests) 0.4 $ illustrative

Message Queue / Stream#

Preset Setting Value Basis
Kafka, core CPU (cpuCores) 4 vCPU illustrative
Memory (memoryGib) 16 GiB illustrative
Storage (storageGib) 1,000 GiB illustrative
Partitions (partitions) 12 illustrative
Consumers (consumers) 6 illustrative
Max backlog (maxDepth) 10,000,000 msgs illustrative
Publish (ack) time (publishTimeMs) lognormal, median 1 ms, p95 4 ms illustrative
Consumer time (consumerTimeMs) lognormal, median 2 ms, p95 8 ms illustrative
Max deliveries (maxDeliveries) 3 illustrative
Brokers (brokers) 3 illustrative
Cost per broker-hour (costPerHour) 0.21 $/h illustrative
RabbitMQ, experimental CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 8 GiB illustrative
Storage (storageGib) 200 GiB illustrative
Partitions (partitions) 1,000 illustrative
Consumers (consumers) 8 illustrative
Max backlog (maxDepth) 10,000,000 msgs illustrative
Publish (ack) time (publishTimeMs) lognormal, median 1 ms, p95 4 ms illustrative
Consumer time (consumerTimeMs) lognormal, median 3 ms, p95 12 ms illustrative
Max deliveries (maxDeliveries) 5 illustrative
Message size (messageSizeKb) 4 KiB illustrative
Brokers (brokers) 1 illustrative
Cost per broker-hour (costPerHour) 0.1 $/h illustrative
SQS-style queue, core Partitions (partitions) 10,000 illustrative
Consumers (consumers) 10 illustrative
Max backlog (maxDepth) none illustrative
Publish (ack) time (publishTimeMs) lognormal, median 15 ms, p95 40 ms illustrative
Consumer time (consumerTimeMs) lognormal, median 5 ms, p95 20 ms illustrative
Max deliveries (maxDeliveries) 5 illustrative
Brokers (brokers) 1 illustrative
Cost per broker-hour (costPerHour) 0 $/h illustrative
Pub/Sub-style topic, experimental Partitions (partitions) 10,000 illustrative
Consumers (consumers) 10 illustrative
Max backlog (maxDepth) none illustrative
Publish (ack) time (publishTimeMs) lognormal, median 20 ms, p95 60 ms illustrative
Consumer time (consumerTimeMs) lognormal, median 5 ms, p95 20 ms illustrative
Max deliveries (maxDeliveries) 5 illustrative
Brokers (brokers) 1 illustrative
Cost per broker-hour (costPerHour) 0 $/h illustrative
NATS JetStream, experimental CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 4 GiB illustrative
Storage (storageGib) 100 GiB illustrative
Partitions (partitions) 1,000 illustrative
Consumers (consumers) 8 illustrative
Max backlog (maxDepth) 1,000,000 msgs illustrative
Publish (ack) time (publishTimeMs) lognormal, median 0.5 ms, p95 2 ms illustrative
Consumer time (consumerTimeMs) lognormal, median 2 ms, p95 8 ms illustrative
Brokers (brokers) 3 illustrative
Cost per broker-hour (costPerHour) 0.08 $/h illustrative
Redis Streams, experimental CPU (cpuCores) 2 vCPU illustrative
Memory (memoryGib) 6.4 GiB illustrative
Partitions (partitions) 1,000 illustrative
Consumers (consumers) 4 illustrative
Max backlog (maxDepth) 1,000,000 msgs illustrative
Publish (ack) time (publishTimeMs) lognormal, median 0.5 ms, p95 2 ms illustrative
Consumer time (consumerTimeMs) lognormal, median 2 ms, p95 8 ms illustrative
Brokers (brokers) 1 illustrative
Cost per broker-hour (costPerHour) 0.1 $/h illustrative

Kubernetes cluster#

Preset Setting Value Basis
Kubernetes cluster, core Nodes (nodes) 3 illustrative
vCPU per node (nodeVcpu) 4 vCPU illustrative
Memory per node (nodeMemoryGib) 16 GiB illustrative
Node provisioning time (nodeProvisionS) 120 s illustrative
Cost per node-hour (costPerHour) 0.19 $/h illustrative
Cluster autoscaler (clusterAutoscaler) yes illustrative
Max nodes (maxNodes) 10 illustrative
EKS-style managed cluster, experimental Nodes (nodes) 3 illustrative
vCPU per node (nodeVcpu) 4 vCPU illustrative
Memory per node (nodeMemoryGib) 16 GiB illustrative
Node provisioning time (nodeProvisionS) 150 s illustrative
Cost per node-hour (costPerHour) 0.192 $/h illustrative
Cluster autoscaler (clusterAutoscaler) yes illustrative
Max nodes (maxNodes) 12 illustrative
GKE-style managed cluster, experimental Nodes (nodes) 3 illustrative
vCPU per node (nodeVcpu) 4 vCPU illustrative
Memory per node (nodeMemoryGib) 16 GiB illustrative
Node provisioning time (nodeProvisionS) 90 s illustrative
Cost per node-hour (costPerHour) 0.17 $/h illustrative
Cluster autoscaler (clusterAutoscaler) yes illustrative
Max nodes (maxNodes) 12 illustrative
AKS-style managed cluster, experimental Nodes (nodes) 3 illustrative
vCPU per node (nodeVcpu) 4 vCPU illustrative
Memory per node (nodeMemoryGib) 16 GiB illustrative
Node provisioning time (nodeProvisionS) 180 s illustrative
Cost per node-hour (costPerHour) 0.19 $/h illustrative
Cluster autoscaler (clusterAutoscaler) yes illustrative
Max nodes (maxNodes) 12 illustrative
k3s (small / edge), experimental Nodes (nodes) 2 illustrative
vCPU per node (nodeVcpu) 2 vCPU illustrative
Memory per node (nodeMemoryGib) 4 GiB illustrative
Node provisioning time (nodeProvisionS) 0 s illustrative
Cost per node-hour (costPerHour) 0.04 $/h illustrative
Cluster autoscaler (clusterAutoscaler) no illustrative
Allocatable share (allocatableShare) 0.85 illustrative
Client Setting Value Basis
Mobile app Link Network latency (one way) (latencyMs) lognormal, median 35 ms, p95 120 ms illustrative
Web browser Link Network latency (one way) (latencyMs) lognormal, median 15 ms, p95 45 ms illustrative
API consumers Link Network latency (one way) (latencyMs) lognormal, median 10 ms, p95 30 ms illustrative

Defaults of a new part#

Settings marked as assumptions (≈ in the inspector) are numbers you assert about your system. A new part starts with these illustrative defaults; replace them with your own measurements. Every run lists the values it used under Assumptions.

Part Setting Default
Load Balancer Overhead (overheadMs) 0.2 ms
Cost per hour (costPerHour) 0.025 $/h
Service / API Servers Service time (serviceTimeMs) lognormal, median 8 ms, p95 25 ms
Memory per request (memoryPerRequestMb) 2 MiB
Memory at idle (memoryBaselineMb) 256 MiB
Replica start-up time (startupS) 20 s
Cold start (coldStartMs) the technology's default (above)
GC pause (gcPauseMs) the technology's default (above)
Requests between GCs (gcEveryRequests) the technology's default (above)
JIT warm-up slowdown (warmupFactor) the technology's default (above)
JIT warm-up time (warmupS) the technology's default (above)
Cost per replica-hour (costPerHour) 0.1 $/h
Cache Hit ratio (hitRatio) 0.8
TTL (ttlSeconds) 60 s
Command time (serviceTimeMs) exponential, mean 0.05 ms
Fork stall per GiB (forkMsPerGib) 10 ms
Item size (itemSizeKb) 1 KiB
Cost per hour (costPerHour) 0.1 $/h
Database Read query time (readServiceTimeMs) lognormal, median 5 ms, p95 20 ms
Write query time (writeServiceTimeMs) lognormal, median 10 ms, p95 40 ms
Throughput limit per shard (throughputLimitRps) none
Log flush time (commitFlushMs) lognormal, median 1 ms, p95 3 ms
Replay cost vs. write time (replicaApplyFactor) 0.5
Replication network delay (replicationDelayMs) lognormal, median 0.5 ms, p95 2 ms
SSTables per cold read (sstablesPerRead) the technology's default (above)
Item size (itemSizeKb) 1 KB
CPU share of query time (cpuShare) 0.5
Hot working set (workingSetGib) none
Disk read time (diskReadMs) lognormal, median 1 ms, p95 4 ms
Data size (dataSizeGib) none
Cost per instance-hour (costPerHour) 0.35 $/h
External API Response time (latencyMs) lognormal, median 120 ms, p95 500 ms
Error rate (errorRate) 0.001
Rate limit (rateLimitRps) none
Concurrency limit (concurrencyLimit) none
Cost per million calls (costPerMillionRequests) 0 $
Message Queue / Stream Publish (ack) time (publishTimeMs) lognormal, median 2 ms, p95 8 ms
Consumer time (consumerTimeMs) lognormal, median 5 ms, p95 20 ms
Replication wait (replicationMs) lognormal, median 2 ms, p95 8 ms
Message size (messageSizeKb) 1 KiB
Cost per broker-hour (costPerHour) 0.15 $/h
Kubernetes cluster Allocatable share (allocatableShare) 0.9
Node provisioning time (nodeProvisionS) 120 s
Image pull time (imagePullS) 8 s
Cost per node-hour (costPerHour) 0.19 $/h
Link Network latency (one way) (latencyMs) 0.5 ms
Workload Baseline load (rps) 100 req/s
Read ratio (readRatio) 0.9
Distinct keys (keyCount) 100,000
Key skew (Zipf s) (keySkew) 0.99

Engine constants#

Numbers built into the sizing, cost and component models. Design review thresholds (for example 80% CPU at the peak) are rules of thumb stated with each finding, not inputs to the simulation.

Area Constant Value Basis Code
Auto-sizing CPU utilisation a part is sized for at its peak 60% illustrative sizing.TARGET_CPU
Auto-sizing CPU at the peak a size below the target's may run at, rather than double the cores 65% illustrative sizing.CPU_STEP_SLACK
Auto-sizing Memory headroom over what the model needs +25% illustrative sizing.MEMORY_HEADROOM
Auto-sizing Storage kept under this fill level 70% illustrative sizing.STORAGE_FILL
Auto-sizing Smallest auto storage 10 GiB illustrative sizing.MIN_STORAGE_GIB
Auto-sizing Largest auto memory of one service replica 64 GiB illustrative sizing.MAX_SERVICE_MEMORY_GIB
Auto-sizing Load balancer throughput per vCPU 5,000 req/s illustrative sizing.PROXY_RPS_PER_VCPU
Auto-sizing Load balancer memory per vCPU 1 GiB illustrative sizing.PROXY_GIB_PER_VCPU
Auto-sizing Broker throughput per vCPU 10,000 messages/s illustrative sizing.BROKER_MSGS_PER_VCPU
Auto-sizing Broker memory per vCPU 2 GiB illustrative sizing.BROKER_GIB_PER_VCPU
Auto-sizing Share of a database's data read often, when no hot working set is set 25% illustrative sizing.DB_HOT_SHARE
Auto-sizing Smallest auto memory of a database instance 1 GiB illustrative sizing.DB_MIN_MEMORY_GIB
Auto-sizing Database memory per connection 9,531,392 bytes documented: aws.amazon.com sizing.DB_BYTES_PER_CONNECTION
Auto-sizing A burstable type's CPU at the base load, at most 80% of its baseline documented: aws.amazon.com pricing.BURST_BASE_SHARE
Auto-sizing A burstable type's CPU at the peak, at most 60% of its vCPUs illustrative pricing.BURST_PEAK_SHARE
Auto-sizing Memory of a VM's operating system 0.25 GiB illustrative sizing.OS_MEMORY_GIB
Auto-sizing Serverless function memory per vCPU 1,769 MB documented: aws.amazon.com sizing.LAMBDA_MB_PER_VCPU
Auto-sizing Backlog a delete-on-ack broker keeps room for 60 minutes of peak publishes illustrative sizing.BACKLOG_S
Auto-sizing Log retention of Kafka-style brokers 168 hours documented: kafka.apache.org sizing.LOG_RETENTION_S
Cost estimate Hours in a month 730 illustrative cost.HOURS_PER_MONTH
Cost estimate Price per vCPU-hour (parts without a price) $0.037 illustrative cost.PRICE_PER_VCPU_HOUR
Cost estimate Price per GiB-hour of memory $0.0028 illustrative cost.PRICE_PER_GIB_HOUR
Cost estimate Price per GiB-month of storage $0.08 illustrative cost.PRICE_PER_STORAGE_GIB_MONTH
Cost estimate Bytes an application load balancer processes per request 10 KB illustrative cost.ALB_KB_PER_REQUEST
Cost estimate Processed bytes per load balancer capacity unit (LCU) 1 GB an hour documented: aws.amazon.com cost.ALB_GB_PER_LCU_HOUR
Cost estimate SQS requests per message 3 (send, receive, delete) documented: aws.amazon.com cost.SQS_REQUESTS_PER_MESSAGE
Cost estimate Share of an implied Kubernetes node allocatable to pods 90% illustrative cost.KUBERNETES_ALLOCATABLE
Database model Share of memory used as buffer cache 75% illustrative database.BUFFER_CACHE_SHARE
Database model DynamoDB read units per partition per second 3,000 documented: aws.amazon.com database.PARTITION_READ_UNITS
Database model DynamoDB write units per partition per second 1,000 documented: aws.amazon.com database.PARTITION_WRITE_UNITS
Database model DynamoDB data per partition 10 GiB illustrative database.PARTITION_GIB
Database model DynamoDB on-demand table: reads per second at first 12,000 documented: aws.amazon.com database.ON_DEMAND_READ_UNITS
Database model DynamoDB on-demand table: writes per second at first 4,000 documented: aws.amazon.com database.ON_DEMAND_WRITE_UNITS
Database model DynamoDB burst capacity banked 300 s of unused capacity documented: aws.amazon.com database.BURST_SECONDS
Cache model Share of a node's memory that holds keys 80% illustrative cache.KEY_MEMORY_SHARE
Limits Most component calls one client request may cause 10,000 (a warning above a tenth) illustrative validation.DEFAULT_MAX_AMPLIFICATION