The cloud stories enterprise IT leaders actually read.
Volume 343, Issue 3 · 9 articles
In microservices, a passing test may only prove compatibility with stale assumptions about upstream services, not production safety.
In a monolith, a passing test suite reasonably predicts deployment success because the same team owns code and tests, but cloud-native architectures split ownership across teams, so the cross-service assumptions behind integration tests often belong to nobody.
When an upstream service renames a field or adds a required header, downstream tests keep passing against outdated mocks and the break surfaces in production.
The author, Sancharini Panda, argues this is a structural problem rather than a discipline problem.
The contributed piece says teams must keep behavioral representations of upstream services current to restore the meaning of a green test.
SANS: 93% run production workloads in cloud, but only 40% call their security advanced; misconfiguration drives 45% of critical incidents.
A SANS Institute survey of 400 professionals finds a persistent cloud security maturity gap: while 93% of organizations run production workloads in the cloud, only 40% consider their security strategy advanced.
Misconfiguration causes 45% of critical incidents, more than double credential theft, and 49% need more than 24 hours to remediate critical gaps.
Although 70% raised security budgets, tool sprawl persists and only 21% use a consolidated CNAPP.
SANS instructor Serge Borso blames fragmented management, and experts recommend policy checks in infrastructure-as-code pipelines, tool consolidation and extending zero trust beyond initial authentication.
Not every workload belongs in the cloud: start with a readiness assessment, map dependencies and pilot non-critical systems first.
The article pushes back on the idea that migrating everything to the cloud is the obvious path.
Variable-traffic web applications, remote collaboration tools and offsite disaster recovery are strong cloud candidates, while legacy monoliths with hardware dependencies, regulated datasets and latency-sensitive computing often fit better on-premises or in hybrid models.
It recommends a cloud readiness assessment covering architecture, data gravity, compliance boundaries and latency, plus application dependency mapping before anything moves.
Teams should pilot non-critical workloads in phases and reconcile actual compute, storage and egress bills before full cutover to avoid later repatriation.
David Linthicum uses Google Cloud's London outage to argue that resilience must be designed, funded, tested and owned by enterprises.
David Linthicum says the Google Cloud power event in the europe-west2 London region showed how many unrelated services, from banks to NHS app features, share the same cloud dependencies.
Hyperscalers provide resilience building blocks such as zones, replication and backups, but not a finished strategy.
Workloads in a single region with one database, one identity path and no tested failover remain fragile.
He urges enterprises to map dependencies (zonal, regional, global, control plane and third party), test restores and failover, and treat resilience as an owned discipline.
CSIRO's serverless sBeacon queries genomic data in place on S3 using Lambda, Athena and DynamoDB, at fractions of a cent per query.
CSIRO built Serverless Beacon (sBeacon), an implementation of the GA4GH Beacon standard that searches genomic variant data without copying it into a central database.
VCF files stay in Amazon S3 while metadata is indexed separately, and Lambda functions use bcftools with HTTP byte-range requests to read only the genomic regions needed.
In testing on chromosome 1 from the 1000 Genomes Project, ingesting a 1.1 GB compressed file took 18 seconds and cost about $0.0005, and a 10,000-base-pair query averaged 1.52 seconds.
It is a useful case study in event-driven, query-in-place data design.
Architecture, not negotiated rates, drives multi-cloud egress costs, and FinOps should be built in at design time.
This InfoQ article applies backend FinOps to multi-cloud systems, arguing that cross-cloud egress charges pile up silently through service placement and communication patterns.
It covers cost visibility with the FinOps FOCUS billing specification, noting that service-level attribution still needs provider-specific fields and ownership metadata.
In a controlled experiment, moving from synchronous cross-cloud processing to near-source processing saved over $2,000 a month, versus $825 from a better provider rate.
The author recommends scheduled synchronization, careful workload placement and CI/CD-integrated cost checks.
Applying FinOps to networking ties egress, WAN and interconnect spending to the teams and business value behind it.
TechTarget argues that network costs have moved from predictable hardware and carrier contracts to consumption-driven charges such as cloud egress, cross-zone traffic, SD-WAN, interconnects and security services.
That makes data movement a financial architecture decision.
Network FinOps uses billing data, telemetry, flow data and contract details to show what was bought and how it was used, then assigns accountability to the teams driving consumption.
It cautions against overly granular reporting, and against assuming the cheapest traffic path is best without weighing latency, availability, security and compliance.
Cost, sovereignty and AI are driving selective cloud repatriation; CISOs must plan security for workloads moving back on-premises.
Cloud repatriation is a deliberate reassessment of workload placement rather than a mass exodus.
A Cloudian and Centiment survey of 212 IT decision-makers found three-quarters had moved some workloads back from public cloud in the past two years, with data sovereignty cited by 81.7%, budget overruns from egress and residency pricing by 84%, and AI latency needs by 85.1%.
Nearly 30% of those repatriating also expanded their cloud footprint.
The article covers which workloads are moving, including data-intensive, business-critical, AI and latency-sensitive apps, and the security and compliance considerations CISOs face in the shift.
Netflix lets Spark jobs on EMR swap their AWS IAM identity for short-lived internal mTLS certificates, with no injected secrets.
Netflix built a workload attestation mechanism so Apache Spark jobs on Amazon EMR can exchange their AWS execution identity for a short-lived certificate from its internal Metatron PKI.
Each Data Project maps to a dedicated IAM role spread across a pool of AWS accounts.
The Spark driver sends a pre-signed STS GetCallerIdentity request, which the identity service checks against signed metadata from the control plane before issuing a certificate.
The driver attests once and passes credentials to executors over encrypted Spark RPC, then re-attests periodically for long-running jobs.