Farid's AngleField notes on data centre infrastructure, AI platforms, virtualization, storage and the cost of running all of it. By Farid, Algiers.https://www.farids.dev/What Live Optics collects, and what it quietly does nothttps://www.farids.dev/articles/live-optics-what-it-collects/https://www.farids.dev/articles/live-optics-what-it-collects/First in a short series on the tool everyone runs before a storage or platform proposal. Knowing the shape of the data is the difference between a sizing and a guess.Sat, 12 Sep 2026 00:00:00 GMTTwenty-four hours of Live Optics is not a workload profilehttps://www.farids.dev/articles/live-optics-collection-window/https://www.farids.dev/articles/live-optics-collection-window/Second in the series. The collection window is the single input that most often invalidates a sizing, and it is the one nobody negotiates.Fri, 11 Sep 2026 00:00:00 GMTReading Live Optics storage numbers without being ledhttps://www.farids.dev/articles/live-optics-reading-storage-numbers/https://www.farids.dev/articles/live-optics-reading-storage-numbers/Third in the series. IOPS, block size, read/write mix and the compression estimate — what each one is actually measuring, and where the sizing goes wrong.Thu, 10 Sep 2026 00:00:00 GMTFrom collection to proposal: write the assumptions on the pagehttps://www.farids.dev/articles/live-optics-to-proposal/https://www.farids.dev/articles/live-optics-to-proposal/Last in the Live Optics series. The arithmetic between the data and the recommendation is where every argument happens, and it is almost always invisible.Wed, 09 Sep 2026 00:00:00 GMTThe order I read a Broadcom renewal quote inhttps://www.farids.dev/articles/vmware-renewal-read-order/https://www.farids.dev/articles/vmware-renewal-read-order/Most people open the VMware renewal at the total. That number is the last thing you should look at, and by then you have already lost the argument.Tue, 08 Sep 2026 00:00:00 GMTWhat a parallel file system actually ishttps://www.farids.dev/articles/what-is-a-parallel-file-system/https://www.farids.dev/articles/what-is-a-parallel-file-system/It is not a faster NAS. The architectural difference is that clients talk to many storage servers at once, and almost every operational surprise follows from that.Mon, 07 Sep 2026 00:00:00 GMTMetadata is the bottleneck, not bandwidthhttps://www.farids.dev/articles/pfs-metadata-is-the-bottleneck/https://www.farids.dev/articles/pfs-metadata-is-the-bottleneck/Every parallel file system is sold on GB/s. Almost every one that disappoints in production is disappointing because of operations per second on small files.Sun, 06 Sep 2026 00:00:00 GMTStripe count, stripe size, and the defaults that cost you most of your bandwidthhttps://www.farids.dev/articles/pfs-striping-defaults/https://www.farids.dev/articles/pfs-striping-defaults/Two numbers control whether your parallel file system behaves like dozens of servers or like one. Most sites never change them.Sat, 05 Sep 2026 00:00:00 GMTLustre in the real world: what operators actually complain abouthttps://www.farids.dev/articles/lustre-in-the-real-world/https://www.farids.dev/articles/lustre-in-the-real-world/The performance is not the issue. The issues are upgrades, client evictions, one slow target poisoning everything, and the fact that it assumes you employ someone who knows it.Fri, 04 Sep 2026 00:00:00 GMTIBM Storage Scale: what it does well, and the licensing conversationhttps://www.farids.dev/articles/storage-scale-gpfs-notes/https://www.farids.dev/articles/storage-scale-gpfs-notes/GPFS by its older name. Technically excellent, operationally more forgiving than Lustre, and sold in a way that requires you to read carefully.Thu, 03 Sep 2026 00:00:00 GMTBeeGFS and the cluster that is too big for NFS and too small for Lustrehttps://www.farids.dev/articles/beegfs-mid-scale/https://www.farids.dev/articles/beegfs-mid-scale/There is a real gap in the middle of this market, and the thing that fills it is usually the one nobody shortlisted.Wed, 02 Sep 2026 00:00:00 GMTAll-flash parallel file systems: what changed and what did nothttps://www.farids.dev/articles/all-flash-parallel-file-systems/https://www.farids.dev/articles/all-flash-parallel-file-systems/Flash removed the seek penalty that shaped twenty years of design. It did not remove the metadata bottleneck, the network limit, or the need to think about layout.Tue, 01 Sep 2026 00:00:00 GMTSizing a parallel file system: start with bandwidth, not terabyteshttps://www.farids.dev/articles/sizing-a-parallel-file-system/https://www.farids.dev/articles/sizing-a-parallel-file-system/Capacity is the easy number and the wrong one to start from. Here is the order that produces a design somebody can defend.Mon, 31 Aug 2026 00:00:00 GMTCheckpointing is the write burst that sizes your file systemhttps://www.farids.dev/articles/checkpoint-write-burst/https://www.farids.dev/articles/checkpoint-write-burst/A cluster that reads steadily and writes rarely still needs to absorb its entire memory footprint in a few minutes. That burst, not the average, decides the design.Sun, 30 Aug 2026 00:00:00 GMTNutanix is not a drop-in for vSphere, and pretending otherwise costs you month fourhttps://www.farids.dev/articles/nutanix-is-not-a-drop-in/https://www.farids.dev/articles/nutanix-is-not-a-drop-in/AHV will run your VMs. That was never the risky part. The risk is in the twenty things around the VM that you forgot were vSphere features.Sat, 29 Aug 2026 00:00:00 GMTGPUDirect Storage: what it buys you, and when it does nothttps://www.farids.dev/articles/gpudirect-storage-reality/https://www.farids.dev/articles/gpudirect-storage-reality/Moving data from storage into accelerator memory without a detour through the host. Real gains in specific conditions, and a long list of prerequisites.Fri, 28 Aug 2026 00:00:00 GMTThe small-file problem, and how machine learning made it worsehttps://www.farids.dev/articles/small-files-problem/https://www.farids.dev/articles/small-files-problem/Every shared file system struggles with millions of tiny files. Training pipelines produce exactly that, and the fix is in the pipeline rather than the storage.Thu, 27 Aug 2026 00:00:00 GMTHow parallel file systems fail: the modes nobody warns you abouthttps://www.farids.dev/articles/pfs-failure-modes/https://www.farids.dev/articles/pfs-failure-modes/They rarely stop. They degrade, partially and confusingly, in ways that present to users as 'the storage is slow' with no pattern.Wed, 26 Aug 2026 00:00:00 GMTYou cannot back up a parallel file system the way you thinkhttps://www.farids.dev/articles/backing-up-a-parallel-file-system/https://www.farids.dev/articles/backing-up-a-parallel-file-system/Walking a namespace of a billion files with a conventional backup agent will not finish. The strategies that work look nothing like enterprise backup.Tue, 25 Aug 2026 00:00:00 GMTSizing a backup appliance: the five numbers that matterhttps://www.farids.dev/articles/backup-appliance-sizing/https://www.farids.dev/articles/backup-appliance-sizing/Front-end capacity, change rate, retention, ingest rate and restore rate. Get these and the configuration follows. Skip one and the appliance is wrong.Mon, 24 Aug 2026 00:00:00 GMTDeduplication ratios are a marketing number until you name the datasethttps://www.farids.dev/articles/dedupe-ratios-are-marketing/https://www.farids.dev/articles/dedupe-ratios-are-marketing/A 50:1 ratio and a 4:1 ratio can both be honest. The difference is what was measured, and the assumption in your quote determines whether the appliance fits.Sun, 23 Aug 2026 00:00:00 GMTNobody sizes for restorehttps://www.farids.dev/articles/nobody-sizes-for-restore/https://www.farids.dev/articles/nobody-sizes-for-restore/Backup windows get modelled to the hour. The restore rate, which is the number that matters on the day something has gone wrong, is usually nowhere in the document.Sat, 22 Aug 2026 00:00:00 GMTThe backup appliance market: who is winning, and at whathttps://www.farids.dev/articles/backup-appliance-landscape/https://www.farids.dev/articles/backup-appliance-landscape/The category split in two. Dedicated backup storage is consolidating around fewer hardware vendors while the growth moved to software platforms that sell cyber resilience.Fri, 21 Aug 2026 00:00:00 GMTParallel file system or scale-out NAS: the decision most teams get backwardshttps://www.farids.dev/articles/pfs-or-scale-out-nas/https://www.farids.dev/articles/pfs-or-scale-out-nas/The default answer for an enterprise AI cluster should be NAS, and the burden of proof should sit with the parallel file system. Usually it is the other way round.Thu, 20 Aug 2026 00:00:00 GMTInference is not training, and your sizing model probably assumes it ishttps://www.farids.dev/articles/inference-is-not-training/https://www.farids.dev/articles/inference-is-not-training/Two workloads with opposite characteristics, sized with the same reference architecture. The result is a cluster that is wrong in both directions at once.Wed, 19 Aug 2026 00:00:00 GMTWhat OpenShift Virtualization quietly asks of your storage teamhttps://www.farids.dev/articles/kubevirt-storage-tax/https://www.farids.dev/articles/kubevirt-storage-tax/KubeVirt runs VMs inside pods. That sentence hides an entire storage conversation that most migration plans discover in the pilot.Tue, 18 Aug 2026 00:00:00 GMTObject storage for AI datasets: cheaper, slower, and often the right answerhttps://www.farids.dev/articles/object-storage-for-ai-datasets/https://www.farids.dev/articles/object-storage-for-ai-datasets/Not everything needs to sit on the fast tier. The interesting design question is what moves between tiers and who decides.Mon, 17 Aug 2026 00:00:00 GMTAn RVTools export is not an assessmenthttps://www.farids.dev/articles/rvtools-is-not-an-assessment/https://www.farids.dev/articles/rvtools-is-not-an-assessment/Everybody starts with the same spreadsheet. Almost nobody reads the twelve columns that change the answer, and every vendor tool quietly fills the gaps in its own favour.Wed, 05 Aug 2026 00:00:00 GMTWhat a GPU node costs before it produces a single tokenhttps://www.farids.dev/articles/cost-before-the-first-token/https://www.farids.dev/articles/cost-before-the-first-token/The accelerators are the headline. Power, cooling, fabric, storage and the eighteen weeks of waiting are the budget.Tue, 28 Jul 2026 00:00:00 GMTNobody sizes the fabric, then blames the GPUshttps://www.farids.dev/articles/nobody-sizes-the-fabric/https://www.farids.dev/articles/nobody-sizes-the-fabric/East-west traffic in an AI cluster behaves nothing like the enterprise traffic your network was designed for. The symptom looks like slow accelerators.Tue, 14 Jul 2026 00:00:00 GMTIndexing 604 pages of recitation with Whisper and a lot of fuzzy matchinghttps://www.farids.dev/articles/indexing-quran-recitations/https://www.farids.dev/articles/indexing-quran-recitations/A weekend project that taught me more about applied AI than any vendor deck: transcription is the easy half, alignment is the real problem, and the human in the loop is not a failure.Tue, 30 Jun 2026 00:00:00 GMTRAG is a retrieval problem wearing an AI costumehttps://www.farids.dev/articles/rag-is-a-retrieval-problem/https://www.farids.dev/articles/rag-is-a-retrieval-problem/Nine out of ten disappointing RAG systems are disappointing search systems. The generation step is rarely where the failure is.Tue, 16 Jun 2026 00:00:00 GMTNVMe/TCP quietly won the mid-range while everyone argued about Fibre Channelhttps://www.farids.dev/articles/nvme-tcp-won-the-midrange/https://www.farids.dev/articles/nvme-tcp-won-the-midrange/Not because it is faster. Because it removed a team, a skill set and a second network from the build sheet.Tue, 26 May 2026 00:00:00 GMTYour immutable backup is not an air gaphttps://www.farids.dev/articles/immutable-is-not-air-gapped/https://www.farids.dev/articles/immutable-is-not-air-gapped/Both words get used as though they mean the same protection. They do not, and the difference is the credential that can reach the repository.Tue, 12 May 2026 00:00:00 GMTThe three questions that end most 'we will just move to the cloud' planshttps://www.farids.dev/articles/cloud-repatriation-arithmetic/https://www.farids.dev/articles/cloud-repatriation-arithmetic/Not arguments against public cloud. Just the three numbers that tend to be missing when the business case looks obviously good.Mon, 27 Apr 2026 00:00:00 GMT