---
title: "Storage"
description: "8 posts about Storage, summarised, each linking to the original."
---

# Storage
> 8 posts about Storage, summarised, each linking to the original.

## Articles

### [Introducing Together Storage Accelerator: fast, free data caching for your GPU cluster](https://yomu.fyi/post/introducing-together-storage-accelerator-fast-free-data-caching-for-yo.md)
- Company: [Together AI](https://yomu.fyi/company/together-ai.md)
- Published: Nov 19, 2026
- Type: Announcement

Together Storage Accelerator (TSA) is a distributed caching layer for Together GPU Clusters that eliminates repeated data reads from object storage and Hugging Face Hub. It runs on every GPU node and presents drop-in endpoints for existing storage, requiring no application changes. The first read from any node caches data on cluster NVMe, so subsequent reads from any node are served locally. Measured on an H100 node, warm reads are about 20 times faster than cross-cloud downloads, reaching up to 3.4 GB/s per node, and a 70 GB checkpoint pulled by 200 nodes downloads from the source only once. TSA is included free with every Together GPU Cluster and is powered by Together’s ASTRA distributed caching technology.


### [Lightbits is now certified for Red Hat OpenStack Services on OpenShift](https://yomu.fyi/post/lightbits-is-now-certified-for-red-hat-openstack-services-on-openshift.md)
- Company: [Red Hat](https://yomu.fyi/company/red-hat.md)
- Author: Gregory Charot
- Published: Nov 10, 2026
- Type: Announcement

Lightbits is now a certified Cinder driver for Red Hat OpenStack Services on OpenShift, tested on release 18.0.22 with driver version 2.3.12. Certification means that if an issue involves the partner component, Red Hat provides standard support and SLA while collaborating with the partner, rather than bouncing the customer back to the storage vendor. The driver ships from upstream OpenStack without forking or patching, and deployment uses standard OpenStackControlPlane custom resources and a discovery client on each compute node. New features covered include boot from volume, create volume from image, snapshot, backup, retype, and multi-attach. Retype is expected to be the most common operation for adopting Lightbits, allowing volume migration across backends using standard Cinder calls.


### [Inside GitHub’s new Git storage architecture](https://yomu.fyi/post/inside-github-s-new-git-storage-architecture.md)
- Company: [Github](https://yomu.fyi/company/github.md)
- Author: Karthik Raman
- Published: Nov 5, 2026
- Type: Announcement

GitHub announced Project Rubicon, a new storage architecture supporting thousands of writes per second and up to 5,000 agents per repo, tested at 7,200 pushes per second. The design minimizes coordination by decoupling storage from compute, leveraging that only reference updates need ordered agreement because Git objects are immutable, and indexing packs is CPU-heavy so writes use multiple nodes. Writes upload packs to Azure Blob Storage and commit a reference change to Ref-CAS, a log storing packs and references. Reads check the reference version with Ref-CAS and download missing packs into a virtual replica’s cache, which can be rebuilt from scratch. A compaction fleet merges packs and removes unreachable objects, and the private preview architecture, after migrating 99.99999% of repositories to Azure, aims to scale agentic development without changing Git client experience.


### [How to run I/O workloads on OpenShift Virtualization VMs](https://yomu.fyi/post/how-to-run-i-o-workloads-on-openshift-virtualization-vms.md)
- Company: [Red Hat](https://yomu.fyi/company/red-hat.md)
- Author: Elvir Kuric
- Published: Nov 3, 2026
- Type: Tutorial

Testing I/O performance at scale across multiple OpenShift Virtualization VMs is challenging due to scattered SSH sessions and ad hoc scripts. The fio-container harness turns multi-VM FIO into a single, repeatable container run using a Python orchestration script. It supports both pre-created VMs and on-demand provisioning (CREATE\_VMS=true) for Linux guests based on Fedora or RHEL. The harness handles storage preparation, FIO installation, data pre-write, and cleanup, and collects per-VM JSON results into a single mounted directory. Results can be analyzed with provided Python scripts to generate IOPS and bandwidth graphs for comparing StorageClass configurations or cluster changes.


### [How to Check and Configure NVIDIA GPUDirect Storage for Your Platform](https://yomu.fyi/post/how-to-check-and-configure-nvidia-gpudirect-storage-for-your-platform.md)
- Company: [NVIDIA](https://yomu.fyi/company/nvidia-developer-blog.md)
- Author: Phil Carns
- Published: Oct 30, 2026
- Type: Tutorial

NVIDIA GPUDirect Storage (GDS) accelerates GPU storage access by providing a direct data path between GPUs and storage, reducing latency and CPU overhead. The gds-diag tool, an open-source command-line utility distributed via the NVIDIA Magnum IO GitHub repository, simplifies configuring and optimizing GDS deployments by assessing environments, identifying problems, and suggesting improvements. It includes an optional AI agent skill that adds a natural-language interface and orchestration layer for multi-step diagnostics, working with coding assistants like Claude and Codex. The tool supports bare-metal, virtual machine, and container environments, and its subcommands include pre-install, post-install, config-audit, mount-check, and container-check, with each finding having a status badge (FAIL, WARN, INFO, PASS) and a mitigation plan. It requires Python 3.8 or newer, does not modify the system, and can output findings as structured JSON, while the support-matrix subcommand shows supported platform combinations with color-coded compatibility.


### [OrioleDB public beta: a Postgres storage engine for update-heavy tables](https://yomu.fyi/post/orioledb-public-beta-a-postgres-storage-engine-for-update-heavy-tables.md)
- Company: [Supabase](https://yomu.fyi/company/supabase.md)
- Author: aditya\_maruvada,daniel\_mitterdorfer
- Published: Oct 28, 2026
- Type: Announcement

OrioleDB, an open-source storage engine for Postgres, is now in public beta on every Supabase plan. It replaces the heap with an undo log, 64-bit transaction IDs, index-organized tables, and lock-free page reads to eliminate dead tuples, reduce bloat, and improve concurrency. Performance benchmarks on DBArena show that when data exceeds memory, OrioleDB achieved up to 1.8x more transactions per minute than heap storage—113,850 on an 8XL instance—and about 46% more throughput per dollar at small sizes. The engine is chosen at project creation, billed the same as standard Postgres, but has limitations such as no Point-in-Time Recovery, unsupported VACUUM FULL, and experimental non-B-tree indexes. The Supabase team recommends benchmarking your own workload, as OrioleDB helps most on update-heavy tables with primary-key lookups, data larger than memory, and high concurrency.


### [Deploying third-party applications to Red Hat OpenShift: Common pitfalls and how to avoid them](https://yomu.fyi/post/deploying-third-party-applications-to-red-hat-openshift-common-pitfall.md)
- Company: [Red Hat](https://yomu.fyi/company/red-hat.md)
- Author: Alasdair Preston
- Published: Oct 27, 2026
- Type: Explainer

Enterprise customers deploying third-party applications on Red Hat OpenShift often encounter vendors claiming Kubernetes support without addressing the underlying architectural changes needed for containerization. These applications are frequently repackaged from legacy virtual machines, retain tight pod interdependencies, and demand cluster-admin privileges or unrestricted security context constraints, violating OpenShift’s default restricted-v2 SCC. Many products fail to support parallel dev, test, and staging environments because they depend on cluster-scoped resources, and they rely on legacy NFS shared storage instead of dynamic provisioning via OpenShift Data Foundation. Avoiding these pitfalls requires formal architectural reviews with vendor engineering, strict enforcement of non-root execution within defined SCC policies, and conversion of manual vendor scripts into declarative Helm charts managed through GitOps. Consultants must also manage expectations with evidence-based feedback and ensure thorough knowledge transfer so the customer’s own engineering team can maintain a clean, repeatable platform.


### [Origin Range Requests: Serving large files without range-request tuning](https://yomu.fyi/post/origin-range-requests-serving-large-files-without-range-request-tuning.md)
- Company: [Cloudflare](https://yomu.fyi/company/cloudflare.md)
- Author: Matthew Gumport
- Published: Oct 23, 2026
- Type: Announcement

Naive byte-range caching in CDNs can waste huge bandwidth when a small client request triggers a large origin fetch—for example, a seek request for a few megabytes could pull 128 MiB. Cloudflare’s previous implementation used fixed range sizes that created tradeoffs; the size was part of the cache key, so updates invalidated entries, and it could not combine with concurrent streaming, forcing downloads before serving. The new Origin Range Requests feature, a self-service option with no tuning, uses Pingora’s fixed-size chunks to align client ranges to chunk boundaries and fetch only the aligned chunk. A coordinator merges bytes from cache, in-flight downloads, and origin responses, checking entity tags and refusing to combine across representation changes. Enabling Origin Range Requests halved p99 time to first byte for R2, cut tiered cache traffic by roughly 6 percent, reducing petabyte-scale daily data movement.
