Aerospike · Hybrid Memory · Smart Cluster · XDRAerospike, sub-millisecond at millions of TPS.
Aerospike is a distributed real-time NoSQL database engineered with a Hybrid Memory Architecture (HMA) that pairs in-DRAM primary indexes with raw SSD block storage. Capable of executing millions of transactions per second at sub-millisecond p99 latency, Aerospike powers ad-tech bidding, fraud detection, and financial telecommunications. JusDB provides Aerospike DBRE: raw NVMe profiling, XDR cross-DC tuning, Strong Consistency (SC) failover, and agreed incident response, with 24x7 on-call available.
We consult on high-performance NoSQL for real-time applications that need sub-millisecond latency. Our Aerospike work covers optimization and deployment.
Aerospike 7 · hybrid-memory
Smart Cluster · 6 nodes · RF=2
1.24M
380µs
64.2%
RF=2 ✓
Cluster Throughput
1.24M TPS[OK] defrag: device /dev/nvme0n1 at 4.2% — block reclaim healthy
[INF] migration: partition rebalance complete, 4096/4096 stable
[OK] xdr: shipping to dc-eu lag 38ms, 0 retransmits
[INF] namespace user_data: high-water-mark 64% of 512G
Example monitoring view showing database activity and engineer actions. Uses sample data, not customer data.
Running Aerospike at production scale?
- ▸ Hybrid Memory tuning — your sub-millisecond p99 latency target is sliding. The DRAM + SSD storage profile needs an audit against real working-set and write-amplification numbers.
- ▸ Hot-key contention — specific partitions throttle at peak load, even though namespace-level resources are unsaturated. Your namespace + set partitioning needs a review.
- ▸ XDR cross-DC replication — lag is climbing past SLA. The conflict-resolution policy is unclear. The multi-region cutover runbook hasn't been validated under load.
JusDB Aerospike specialists ship instrumented tuning runbooks for production clusters at scale. See Aerospike consulting →
24
Supported database technologies
24x7
On-call coverage, under your service agreement
30 min
P1 initial-response target on 24x7 on-call
2 to 4 h
Initial-response target in covered business hours
Aerospike engineering
Our Aerospike specialists tune, scale, and secure clusters for demanding real-time applications.
Latency Optimization
Achieve strict sub-millisecond latencies through Hybrid Memory tuning and SSD profiling.
Cluster Scaling & Sizing
Capacity planning and hot-resizing of Aerospike clusters for high-throughput peaks.
Aerospike Connect Integration
Stream data efficiently to Kafka, Spark, and Presto for downstream real-time analytics.
Enterprise Hardening
Configure Role-Based Access Control (RBAC), LDAP, and TLS/mTLS encryption for compliance.
Observability
Configure Prometheus exporters and Grafana for early detection of SSD wear and CPU bottlenecks.
Cross-DC Replication (XDR)
Configure XDR for multi-region failover and distributed edge-locality reads.
Which Aerospike service do you actually need?
Match the symptom you're seeing to the engagement that fixes it. If more than one row matches, start with consulting. It settles the architecture, and most other engagements depend on the topology being right.
| Symptom / situation | Right engagement | Why |
|---|---|---|
| Moving from DynamoDB / Cassandra / Redis to Aerospike | Aerospike Migration | Cutover playbook, dual-write topology, data-model translation, rollback gates. |
| Sub-millisecond p99 latency is sliding under load | Performance Tuning | Hybrid Memory + SSD storage profile audit, namespace tuning, hot-key remediation, write-amplification analysis. |
| Production cluster down or replication-lag incident | Aerospike Support | Agreed incident response (24x7 on-call available), root-cause analysis, post-mortem playbooks. |
| Multi-region active-active design / XDR cutover | High Availability | XDR cross-DC replication topology, conflict-resolution policy, region-failover runbook. |
| Ongoing operations — no in-house Aerospike DBA | Remote DBA | Dedicated DBA-as-a-service: monitoring, patching, capacity planning, query tuning. |
| Aerospike on Kubernetes / EKS / GKE | Aerospike on Kubernetes | Aerospike Kubernetes Operator (AKO), StatefulSet design, persistent storage class, multi-AZ topology. |
| Architecture decision before any code ships | Aerospike Consulting | Topology design, namespace + set strategy, hybrid memory sizing, DynamoDB / Cassandra comparison. |
Aerospike at a glance — what the leaves don't cover
Aerospike is a NoSQL key-value + document database built around Hybrid Memory Architecture. Indexes sit in DRAM. Data sits on SSD with raw block access, bypassing the filesystem. That single architectural choice is why Aerospike consistently delivers sub-millisecond p99 latency at scale. It holds that latency where Redis runs out of RAM budget and Cassandra chokes on GC pauses. So it becomes the database of record for real-time fraud detection, ad-tech bidstreams, gaming leaderboards, and telecom subscriber profiles. In those workloads the latency budget is measured in microseconds, and you cannot afford a stop-the-world GC.
Three trade-offs are worth knowing before you adopt it:
- ▸ Strong Consistency (SC) vs Available (AP) — the two modes have different semantics. The one you pick changes failover behaviour.
- ▸ XDR replication is always asynchronous — Cross-Datacenter Replication needs operational discipline to get right under load.
- ▸ Enterprise is rarely optional — XDR, RBAC, on-disk encryption, and Aerospike Connect for Kafka/Spark all ship in Aerospike Enterprise (paid). Production deployments end up needing them, so the "Community is enough" assumption rarely survives the first scaling event.
Aerospike expertise
at microsecond latency
We tune defrag thresholds and size the primary index in DRAM. We profile SSD write-blocks and audit XDR ship throughput. The result is a cluster that holds sub-millisecond p99 under millions of transactions per second.
Cluster Performance
After tuningIllustrative health scores · results depend on the workload
Workloads we've transformed
Illustrative examples of the problem and the fix, not client measurements.
1,800µs
210µs
Redis cache layer + separate datastore
The fix
Collapsed cache + store into Aerospike hybrid-memory — 1 hop, ~70% lower infra cost
9,400µs
340µs
Single hot key serializing writes
The fix
Sharded key + write-block tuning; p99 dropped below 0.5ms
6,200µs
180µs
Secondary-index query scanning SSD
The fix
Sized primary + secondary index fully in RAM; data on NVMe
4/4
Nodes stable
18s
Rebalance Complete
420µs
p99 Read Latency
Always on. Shared-nothing.
Aerospike's Smart Cluster is shared-nothing, with no master. Partitions distribute themselves across nodes and rebalance on every membership change. XDR ships data cross-datacenter for multi-region failover. A lost node recovers in seconds on a topology designed for the availability target agreed for your workload.
A latency P1, answered within your agreement's target.
When defrag falls behind on NVMe, the write-block queue backs up. A named Aerospike engineer responds to that, not a ticket queue. Raising defrag thresholds and sprigs, then a rolling node restart, caught the cluster up with no downtime. A blameless postmortem followed.
p99 read latency rising under namespace pressure
A named Aerospike engineer responds within your agreement's initial-response target
Defrag falling behind on NVMe; write-block queue backing up
Defrag thresholds and write-block settings tuned, then a rolling node restart
Write-block queue drained and p99 back to baseline; capacity review and post-incident review
Pre-Migration Assessment
Redis / Cassandra / RDBMS → Aerospike
Cutover via dual-write + bulk loader: near-zero downtime
Move to Aerospike with minimal downtime
Redis, Cassandra, DynamoDB or an RDBMS → Aerospike. First we translate the data model and key mapping. Then we bulk-load with asbackup or the loader. Dual-write replication catches the new cluster up, and we cut over in a near-zero-downtime window.
Specialized Aerospike Services
Aerospike engagements we run most often. Pick the one that matches the work in front of you.
Aerospike Consulting
Expert guidance on architecture, implementation, and best practices for your Aerospike deployment.
Performance Tuning
Optimize your Aerospike cluster for maximum throughput and minimum latency.
Aerospike Migration
Low-downtime migration to Aerospike from other databases, with rehearsed cutovers.
Aerospike Support
Agreed expert support and maintenance for your mission-critical Aerospike systems, with 24x7 on-call available.
High Availability
Design and implement highly available Aerospike architectures with automatic failover.
Remote DBA Services
Dedicated remote database administration for your Aerospike infrastructure.
Aerospike on Kubernetes
AKO operator deployment, StatefulSet design, persistent storage, and multi-AZ topology on EKS or GKE.
Aerospike Production Failure Modes
Sub-millisecond NoSQL systems face critical operational threats from NVMe write amplification, XDR replication lag overflows, and Strong Consistency partition deadlocks. Here is how JusDB DBREs diagnose and eliminate Aerospike's most severe failure modes.
Write Amplification & Raw NVMe Defragmentation Lag
High-throughput write/update bursts overwhelm storage defragmentation threads. When defrag cannot keep pace with continuous block updates, write queues fill up and trigger sudden disk write halts.
Benchmark raw SSD storage with the Aerospike act tool, tune defrag-lwm-pct and defrag-sleep parameters, and deploy multiple disk worker threads across direct raw NVMe block devices.
XDR Buffer Saturation & Cross-Regional Drift
Inter-region WAN latency spikes or remote cluster processing backlogs fill Cross-Datacenter Replication (XDR) memory queues, causing unacknowledged record queues to overflow and inducing split-state drift.
Implement dynamic ship-bin filtering to drop ephemeral tracking counters, configure adaptive xdr-max-throughput limits, and set real-time alerting on xdr_unacked_records and lag metrics.
Strong Consistency (SC) Deadlock & Partition Quorum Loss
In Strong Consistency namespaces, transient network flapping between rack-aware zones trips automatic write deadlocks (split-brain protection), rejecting client writes until manual roster intervention.
Tune auto-dun and cluster-stable-timer settings, enforce odd-count availability zone rack-awareness, and establish non-disruptive roster re-clustering protocols during rolling node maintenance.
Production Aerospike Diagnostic Runbooks
Non-blocking asadm administration commands and asinfo node introspection queries executed by JusDB DBREs during incident triage to isolate defrag latency, XDR replication queue backlogs, and SSD IOPS limits.
Inspects namespace disk free percentages, DRAM index allocations, and defrag queue depth across cluster nodes.
# Query namespace disk free percentage, memory usage, and defrag queue depth asadm -e "show stat namespace like 'used-bytes-disk|used-bytes-memory|defrag-q|free-pct-disk'" # Inspect low-water-mark defrag status and tombstone record count asinfo -v "namespace/prod_kv" -l | grep -E "defrag|write_q|tombstones"
Monitors Cross-Datacenter Replication (XDR) shipped records, unacknowledged record backlogs, and link latency.
# Monitor XDR shipped records, unacknowledged queue backlog, and latency asadm -e "show stat xdr like 'records_shipped|unacked_records|latency_ms'" # Validate XDR active remote datacenter connectivity and dynamic bin filters asinfo -v "get-config:context=xdr" -l
Aerospike DBRE: Evaluation Matrix
How JusDB specialized Aerospike database reliability engineering compares against Aerospike Cloud default managed tier and in-house generalists.
| Evaluation Vector | JusDB Aerospike DBRE | Aerospike Cloud / Default | In-House Generalists |
|---|---|---|---|
| Hybrid Memory Architecture (HMA) & SSD Profiling | Raw block NVMe optimization, act-tool benchmark profiling, DRAM primary index sizing, and write-amplification reduction | Generic instance provisioning without drive stress testing or write queue buffer calibration | File-system overhead on SSDs causing IOPS bottlenecks, cache thrashing, and elevated write latency under sustained load |
| Cross-Datacenter Replication (XDR) & Latency Budget | Active-active multi-region mesh, dynamic ship-bin filtering, unacknowledged write lag monitoring, and failover runbooks tested against your recovery objectives | Basic XDR link configuration without granular bin filtering or WAN bandwidth throttling controls | XDR transaction queues filling up during network degradations, leading to out-of-sync regional states and data loss |
| Strong Consistency (SC) vs AP Mode Partitioning | Namespace-level split-brain protection, linearizable read/write tuning, and roster validation across rack-aware topologies | Default AP mode deployments without formal split-brain protection or cluster roster verification | Accidentally running financial ledgers in AP mode or experiencing partition unavailability due to incorrect roster definitions |
| Secondary Indexing & Set Partition Design | DRAM/Flash secondary index memory modeling, predicate query optimization, and hot-partition key distribution analysis | Basic secondary index creation; no profiling for high-cardinality index memory footprint | Unindexed predicate scans or hot-key partitions exhausting single-core worker threads during peak real-time bidding |
| Aerospike Kubernetes Operator (AKO) & Sizing | Production AKO CRD deployment, persistent volume claim (PVC) direct NVMe binding, graceful rolling upgrades, and rack-aware pod spread | Standard cloud Kubernetes templates requiring manual cluster migration and stateful storage reconfiguration | StatefulSet pod restarts causing abrupt data rebalancing and migration storm that overloads disk I/O |
| Incident Response & Escalation | Named Aerospike engineers join your incident bridge. Initial-response targets: 2 to 4 hours in covered business hours; 30 minutes for P1 incidents with 24x7 on-call. Targets, not resolution guarantees. | Standard cloud support tickets with 1-to-2 hour initial response windows and generalist infrastructure reps | Engineers reading server log hex codes and defrag thresholds at midnight during latency-spike customer outages |
Aerospike consulting & deployment — frequently asked questions
Common questions about Aerospike consulting services
What makes Aerospike different from other NoSQL databases?
Aerospike's hybrid memory architecture pairs DRAM with SSD storage. That is how it holds sub-millisecond latency at massive scale. Aerospike also supports two modes: Strong Consistency (SC) and high-availability (AP). You set the mode per namespace, so each dataset gets either linearizable consistency or maximum availability. That makes it a strong fit for real-time applications.
How do you ensure data consistency in Aerospike clusters?
We set up Aerospike's strong consistency model with clear conflict-resolution policies. Replica synchronization and read/write policies keep data integrity across nodes in the cluster. Where you need multi-region coverage, we add cross-datacenter replication (XDR) on top.
What's the typical performance improvement with Aerospike?
It depends on the use case and the database you are moving from. Aerospike is built for sub-millisecond reads and writes at high throughput, so latency-bound workloads moving off disk-based stores tend to see the largest change. We benchmark your workload on the current system and on Aerospike before cutover, so the gain is measured rather than estimated.
Do you provide ongoing support after implementation?
Yes. Coverage runs in business hours, and 24x7 on-call is available under your service agreement. Support covers monitoring, maintenance, performance tuning, and emergency response, from basic monitoring up to dedicated DBA services.
Can Aerospike handle both OLTP and OLAP workloads?
Aerospike is built for real-time OLTP workloads. For OLAP, we usually recommend a hybrid architecture. Aerospike handles the real-time operations. A companion engine such as Apache Spark or ClickHouse runs the analytical workloads.
What's the migration process from existing databases?
We migrate in phases: assessment, data modeling, parallel running, gradual cutover, and validation. We design cutovers to keep downtime short and agree any planned window with you in advance; we do not guarantee zero downtime.
Technologies We Work With
Complete Aerospike ecosystem and integration tools
Ready for Sub-Millisecond Performance?
Put Aerospike's high-performance NoSQL database behind your real-time applications. Our experts help you reach the speed and scale you need.
DynamoDB to Aerospike migration walkthrough
DynamoDB to Aerospike: The Cost and Performance Optimization Journey
Explore data-model changes, workload sizing, cost assumptions, and dual-write cutover planning for a move from DynamoDB to Aerospike. Use the walkthrough to prepare questions for your own migration assessment.