<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title/><link>/blog/</link><description>Recent content on</description><language>en-us</language><lastBuildDate>Wed, 26 Aug 2026 09:00:00 +0000</lastBuildDate><atom:link href="/blog/" rel="self" type="application/rss+xml"/><item><title>Introducing ClickHouse® on Exoscale: Real-Time Analytics</title><link>/blog/managed-clickhouse-launch/</link><guid>/blog/managed-clickhouse-launch/</guid><pubDate>Wed, 26 Aug 2026 09:00:00 +0000</pubDate><description>Run sub-second analytics on billions of rows in European zones, with ClickHouse 25.8 LTS, generally available vector search and per-second billing.</description></item><item><title>Self-Hosted LLMs: Benefits, Drawbacks, and Infrastructure</title><link>/blog/self-hosted-llm/</link><guid>/blog/self-hosted-llm/</guid><pubDate>Mon, 24 Aug 2026 09:00:00 +0000</pubDate><description>From data control to infrastructure choices, this guide explains what self-hosted LLMs involve and what organizations should consider.</description></item><item><title>GPU Partitioning with MIG on Exoscale SKS</title><link>/blog/share-gpu-using-mig/</link><guid>/blog/share-gpu-using-mig/</guid><pubDate>Thu, 02 Jul 2026 09:00:00 +0000</pubDate><description>&lt;p&gt;GPUs are massively used in the age of AI, but depending on the workload we need to run, a single GPU can sometimes be over-dimensioned.&lt;/p&gt;
&lt;p&gt;Imagine deploying a 7B model on a 48 GB GPU. More than half of the GPU would remain unused. MIG allows several independent inference services to share the same physical GPU without interfering with one another.&lt;/p&gt;
&lt;p&gt;In this post, we&amp;rsquo;ll explore how &lt;strong&gt;MIG (Multi-Instance GPU)&lt;/strong&gt; can partition a physical GPU into isolated slices, each with its own dedicated memory, compute, and bandwidth.&lt;/p&gt;</description></item><item><title>Inside an LLM: From Prompt to Tokens</title><link>/blog/llm-from-prompt-to-tokens/</link><guid>/blog/llm-from-prompt-to-tokens/</guid><pubDate>Mon, 15 Jun 2026 06:00:00 +0000</pubDate><description>Learn how LLM inference transforms a prompt into tokens through prefill and decode, hidden states, attention, MLP, KV cache, and transformer layers.</description></item><item><title>Scaling Prometheus with Thanos: Long-Term Storage, HA, and Global Queries</title><link>/blog/prometheus-limits-at-scale/</link><guid>/blog/prometheus-limits-at-scale/</guid><pubDate>Wed, 10 Jun 2026 12:00:00 +0000</pubDate><description>Prometheus hits scaling limits: local disk caps retention, high cardinality pressures RAM, and there is no global view across clusters. Here is how Thanos addresses each of these limits on Exoscale.</description></item><item><title>From Commercial API to Sovereign Infrastructure: A Practical Guide to LLM Workload Migration</title><link>/blog/sovereign-llm-migration/</link><guid>/blog/sovereign-llm-migration/</guid><pubDate>Wed, 20 May 2026 09:00:00 +0000</pubDate><description>&lt;p&gt;In January 2026, a software company received a routine notification from their cloud AI provider: the model powering their production pipeline was being deprecated. No migration path, no extended support window, no negotiation. The model that their extraction logic, prompts, and evaluation suite had been built around was simply going away, and the &amp;ldquo;newer alternatives&amp;rdquo; were not producing satisfactory results.&lt;/p&gt;
&lt;p&gt;At this point, this story is a predictable one. The risk scenario should have been assessed and planned for.&lt;/p&gt;</description></item><item><title>The 9000-to-1500 cliff: an IPv6 PMTUD field note</title><link>/blog/9000-to-1500-cliff/</link><guid>/blog/9000-to-1500-cliff/</guid><pubDate>Mon, 04 May 2026 07:00:00 +0000</pubDate><description>How a 9000-byte first hop, ECMP, and missing IPv6 TCP MSS clamping caused large TCP responses to stall.</description></item><item><title>Platform Engineering on Exoscale: Building an Internal Developer Platform</title><link>/blog/building-an-idp/</link><guid>/blog/building-an-idp/</guid><pubDate>Mon, 27 Apr 2026 08:00:00 +0000</pubDate><description>Learn how Gravitek built an Internal Developer Platform on Exoscale: Platform engineering focused on autoscaling, security, and self-service.</description></item><item><title>The Best Discord Alternatives for Companies</title><link>/blog/discord-alternatives/</link><guid>/blog/discord-alternatives/</guid><pubDate>Wed, 08 Apr 2026 08:00:00 +0000</pubDate><description>Discover good Discord alternatives, from websites, platforms, messaging apps, software, to chat sites like Discord. Read our guide now.</description></item><item><title>What is Object Storage?</title><link>/blog/what-is-object-storage/</link><guid>/blog/what-is-object-storage/</guid><pubDate>Tue, 07 Apr 2026 09:00:00 +0000</pubDate><description>Object-based Storage explained: Discover how it works and which advantages and disadvantages organizations should know about.</description></item><item><title>How to deploy Matrix on Kubernetes: A complete guide</title><link>/blog/matrix-on-exoscale-sks/</link><guid>/blog/matrix-on-exoscale-sks/</guid><pubDate>Fri, 13 Mar 2026 09:00:00 +0000</pubDate><description>Learn to deploy production-grade Matrix on Exoscale Kubernetes, supporting EU data sovereignty and GDPR compliance.</description></item><item><title>Kubernetes Audit Log available on Exoscale SKS</title><link>/blog/audit-logs-in-sks/</link><guid>/blog/audit-logs-in-sks/</guid><pubDate>Wed, 07 Jan 2026 08:00:00 +0000</pubDate><description>You can now send Exoscale Managed Kubernetes (SKS) audit logs to the webhooks of your choice</description></item><item><title>Sovereign Cloud And Data Sovereignty: An Overview</title><link>/blog/data-sovereignty/</link><guid>/blog/data-sovereignty/</guid><pubDate>Mon, 05 Jan 2026 08:00:00 +0000</pubDate><description>Understand data sovereignty and sovereign cloud: Key definitions, benefits, and practical steps for compliant cloud strategies.</description></item><item><title>Safe, Verifiable AI Code Generation for Mission-Critical Systems</title><link>/blog/scala-business/</link><guid>/blog/scala-business/</guid><pubDate>Tue, 02 Dec 2025 10:00:00 +0000</pubDate><description>Accelerate AI innovation in regulated industries with verifiable, compliant, and high-performance solutions—where speed meets trust and sovereignty.</description></item><item><title>Best European Cloud Provider for Sovereign Hosting</title><link>/blog/european-cloud-providers/</link><guid>/blog/european-cloud-providers/</guid><pubDate>Thu, 27 Nov 2025 09:00:00 +0000</pubDate><description>Comparison of the 10 most relevant European cloud providers in 2026, ranked by data sovereignty, services, pricing, and workload fit.</description></item><item><title>AI Meets Formal Methods: Deploying HES-SO’s Verified Code Pipeline on Exoscale</title><link>/blog/scala-technical/</link><guid>/blog/scala-technical/</guid><pubDate>Fri, 21 Nov 2025 10:00:00 +0000</pubDate><description>Proactive Code Security: Combining Scala’s Type System, Formal Verification, and Exoscale’s Cloud Infrastructure</description></item><item><title>Exoscale’s New Karpenter Integration: Simplifying Kubernetes Node Management</title><link>/blog/sks-karpenter/</link><guid>/blog/sks-karpenter/</guid><pubDate>Mon, 10 Nov 2025 10:00:00 +0000</pubDate><description>&lt;div style="background-color: #f3f4f6; border-radius: 8px; padding: 2rem 2.5rem; margin-top: 2.5rem; margin-bottom: 2.5rem;"&gt;
&lt;h3 style="text-align: center; font-size: 1.25rem; font-weight: 500; color: #6b7280; margin: 0 0 1.5rem 0; letter-spacing: 0.01em;"&gt;Quick summary of this article&lt;/h3&gt;
&lt;ul style="margin: 0; padding: 0; list-style: none; display: flex; flex-direction: column; gap: 1rem;"&gt;
&lt;li style="display: flex; align-items: flex-start; gap: 0.85rem;"&gt;
&lt;svg style="flex-shrink: 0; margin-top: 1px;" width="22" height="22" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg"&gt;
&lt;circle cx="12" cy="12" r="11" stroke="#374151" stroke-width="1.5"/&gt;
&lt;path d="M7 12.5l3.5 3.5 6.5-7" stroke="#374151" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round"/&gt;
&lt;/svg&gt;
&lt;span style="color: #1e3a6e; font-size: 0.95rem; line-height: 1.6;"&gt;Karpenter enables faster, fine-grained autoscaling than Cluster Autoscaler, improving resource utilization across workloads.&lt;/span&gt;
&lt;/li&gt;
&lt;li style="display: flex; align-items: flex-start; gap: 0.85rem;"&gt;
&lt;svg style="flex-shrink: 0; margin-top: 1px;" width="22" height="22" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg"&gt;
&lt;circle cx="12" cy="12" r="11" stroke="#374151" stroke-width="1.5"/&gt;
&lt;path d="M7 12.5l3.5 3.5 6.5-7" stroke="#374151" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round"/&gt;
&lt;/svg&gt;
&lt;span style="color: #1e3a6e; font-size: 0.95rem; line-height: 1.6;"&gt;Exoscale SKS Pro offers Karpenter as a free managed add-on with prerequisites preconfigured by the platform.&lt;/span&gt;
&lt;/li&gt;
&lt;li style="display: flex; align-items: flex-start; gap: 0.85rem;"&gt;
&lt;svg style="flex-shrink: 0; margin-top: 1px;" width="22" height="22" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg"&gt;
&lt;circle cx="12" cy="12" r="11" stroke="#374151" stroke-width="1.5"/&gt;
&lt;path d="M7 12.5l3.5 3.5 6.5-7" stroke="#374151" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round"/&gt;
&lt;/svg&gt;
&lt;span style="color: #1e3a6e; font-size: 0.95rem; line-height: 1.6;"&gt;Integration requires creating ExoscaleNodeClass and NodePool resources to define node templates, scaling policies, and constraints.&lt;/span&gt;
&lt;/li&gt;
&lt;li style="display: flex; align-items: flex-start; gap: 0.85rem;"&gt;
&lt;svg style="flex-shrink: 0; margin-top: 1px;" width="22" height="22" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg"&gt;
&lt;circle cx="12" cy="12" r="11" stroke="#374151" stroke-width="1.5"/&gt;
&lt;path d="M7 12.5l3.5 3.5 6.5-7" stroke="#374151" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round"/&gt;
&lt;/svg&gt;
&lt;span style="color: #1e3a6e; font-size: 0.95rem; line-height: 1.6;"&gt;Karpenter supports heterogeneous instance types including GPUs and spot capacity while enabling aggressive consolidation for cost optimization.&lt;/span&gt;
&lt;/li&gt;
&lt;li style="display: flex; align-items: flex-start; gap: 0.85rem;"&gt;
&lt;svg style="flex-shrink: 0; margin-top: 1px;" width="22" height="22" viewBox="0 0 24 24" fill="none" xmlns="http://www.w3.org/2000/svg"&gt;
&lt;circle cx="12" cy="12" r="11" stroke="#374151" stroke-width="1.5"/&gt;
&lt;path d="M7 12.5l3.5 3.5 6.5-7" stroke="#374151" stroke-width="1.5" stroke-linecap="round" stroke-linejoin="round"/&gt;
&lt;/svg&gt;
&lt;span style="color: #1e3a6e; font-size: 0.95rem; line-height: 1.6;"&gt;Follow best practices by defining multiple ExoscaleNodeClass types and setting accurate pod resource requests and limits.&lt;/span&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;/div&gt;
&lt;p&gt;Exoscale Scalable Kubernetes Service (SKS) has long time support for
&lt;a href="https://github.com/kubernetes/autoscaler"target="_blank" rel="noopener"&gt;cluster-autoscaler&lt;/a&gt; to automatically adjust the number
of nodes in a Kubernetes cluster based on the resource requests of the workloads running in the cluster.
This integration has been a key feature for our users, allowing them to optimize their resource
usage and reduce costs. This implementation relies on SKS Nodepool implementation.&lt;/p&gt;</description></item><item><title>CLOUD Act vs. GDPR: The Conflict About Data Access Explained</title><link>/blog/cloudact-vs-gdpr/</link><guid>/blog/cloudact-vs-gdpr/</guid><pubDate>Tue, 21 Oct 2025 08:00:00 +0000</pubDate><description>Understand the conflict between U.S. CLOUD Act and GDPR: Key risks, safeguards, and practical steps to stay compliant.</description></item><item><title>Your Next Innovation Is Already in Your Data</title><link>/blog/kafka-exoscale-managed-for-innovation/</link><guid>/blog/kafka-exoscale-managed-for-innovation/</guid><pubDate>Wed, 01 Oct 2025 10:00:00 +0000</pubDate><description>Managed Apache Kafka: Your future-proof foundation for data in motion. Achieve scalability, resilience, architectural freedom, and EU data privacy compliance with Exoscale.</description></item><item><title>From 24 Hours to 24 Milliseconds: The Proof of a Real-Time Business</title><link>/blog/kafka-managed-proof/</link><guid>/blog/kafka-managed-proof/</guid><pubDate>Tue, 30 Sep 2025 10:00:00 +0000</pubDate><description>Learn how Exoscale managed Apache Kafka enables real-time fraud prevention, instant e-commerce personalization, and IoT insights. Transform your business with data in motion &amp; boost competitive advantage.</description></item></channel></rss>