080
What Live Optics collects, and what it quietly does not
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.
Articles
Everything published so far, newest first.
079
Twenty-four hours of Live Optics is not a workload profile
Second in the series. The collection window is the single input that most often invalidates a sizing, and it is the one nobody negotiates.
078
Reading Live Optics storage numbers without being led
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.
077
From collection to proposal: write the assumptions on the page
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.
076
The order I read a Broadcom renewal quote in
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.
075
What a parallel file system actually is
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.
074
Metadata is the bottleneck, not bandwidth
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.
073
Stripe count, stripe size, and the defaults that cost you most of your bandwidth
Two numbers control whether your parallel file system behaves like dozens of servers or like one. Most sites never change them.
072
Lustre in the real world: what operators actually complain about
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.
071
IBM Storage Scale: what it does well, and the licensing conversation
GPFS by its older name. Technically excellent, operationally more forgiving than Lustre, and sold in a way that requires you to read carefully.
070
BeeGFS and the cluster that is too big for NFS and too small for Lustre
There is a real gap in the middle of this market, and the thing that fills it is usually the one nobody shortlisted.
069
All-flash parallel file systems: what changed and what did not
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.
068
Sizing a parallel file system: start with bandwidth, not terabytes
Capacity is the easy number and the wrong one to start from. Here is the order that produces a design somebody can defend.
067
Checkpointing is the write burst that sizes your file system
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.
066
Nutanix is not a drop-in for vSphere, and pretending otherwise costs you month four
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.
065
GPUDirect Storage: what it buys you, and when it does not
Moving data from storage into accelerator memory without a detour through the host. Real gains in specific conditions, and a long list of prerequisites.
064
The small-file problem, and how machine learning made it worse
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.
063
How parallel file systems fail: the modes nobody warns you about
They rarely stop. They degrade, partially and confusingly, in ways that present to users as 'the storage is slow' with no pattern.
062
You cannot back up a parallel file system the way you think
Walking a namespace of a billion files with a conventional backup agent will not finish. The strategies that work look nothing like enterprise backup.
061
Sizing a backup appliance: the five numbers that matter
Front-end capacity, change rate, retention, ingest rate and restore rate. Get these and the configuration follows. Skip one and the appliance is wrong.
060
Deduplication ratios are a marketing number until you name the dataset
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.
059
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.
058
The backup appliance market: who is winning, and at what
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.
057
Parallel file system or scale-out NAS: the decision most teams get backwards
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.
056
Inference is not training, and your sizing model probably assumes it is
Two workloads with opposite characteristics, sized with the same reference architecture. The result is a cluster that is wrong in both directions at once.
055
What OpenShift Virtualization quietly asks of your storage team
KubeVirt runs VMs inside pods. That sentence hides an entire storage conversation that most migration plans discover in the pilot.
054
Object storage for AI datasets: cheaper, slower, and often the right answer
Not everything needs to sit on the fast tier. The interesting design question is what moves between tiers and who decides.
053
OpenShift for people who run virtual machines
A translation guide. Most of what you know transfers; the vocabulary and the failure model do not, and the vocabulary is what makes the documentation unreadable at first.
052
Day two Kubernetes: the upgrades are the job
Standing a cluster up is a weekend. Keeping it current for five years across a moving API surface, operator versions and node images is the actual commitment.
051
Operators: the good, the abandoned, and the one you will end up writing
Packaged operational knowledge is the best idea in the Kubernetes ecosystem and the largest source of unmanaged dependency risk in most clusters.
050
etcd will tell you the truth about your cluster
Every Kubernetes component reads from one distributed key-value store. When the cluster is sick, that store is usually why, and almost nobody monitors it.
049
Bare metal or virtualized Kubernetes: choose on the operating model
The performance argument is mostly settled and mostly irrelevant. The question is who does the lifecycle work and what you do on a Tuesday afternoon.
048
GitOps for infrastructure people
The idea is smaller than the vocabulary suggests: the repository is the desired state, and something continuously makes reality match it. What it costs is discipline about where truth lives.
047
The platform team nobody staffed
Kubernetes moves work rather than removing it. The work lands on a team that was not created, and the symptoms look technical.
046
Ingress, load balancers, and the gap between cloud Kubernetes and yours
In a managed cloud cluster, asking for a load balancer produces one. On your own hardware, nothing happens, and the documentation does not mention it.
045
Hybrid is not a strategy, it is a description
Almost every enterprise is hybrid. Saying so commits you to nothing. The strategy is the placement rule, and most organisations do not have one.
044
Committing to cloud capacity without regretting it
Discounts for commitment are real money and a real trap. The question is not how much to commit but how confident you are in a forecast you have never tested.
043
Cloud exit costs, computed properly
You are probably not leaving. Compute it anyway, because a number you cannot calculate is a negotiating position you do not have.
042
Colocation is the third option nobody prices
The debate is framed as our data centre versus public cloud. There is a middle that solves the facilities problem without the consumption bill, and it is increasingly the answer for accelerated infrastructure.
041
An RVTools export 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.
040
FinOps for infrastructure teams who did not ask for it
The bill became your problem the day consumption pricing arrived. Four practices cover most of the value, and none of them require adopting a framework.
039
Disaster recovery in the cloud: the bill is in the test
Standby infrastructure looks cheap until you calculate what it costs to actually fail over, and untested failover is not a capability.
038
Building a private knowledge base for work you cannot paste into a chatbox
Product specifications, caveats and unsupported configurations are exactly what a language model is good at retrieving, and exactly what you cannot send anywhere.
037
Local models on a laptop: what is actually usable
Not a benchmark post. What memory you need, what quantisation costs you, and which tasks a model running on your own machine is genuinely good enough for.
036
When a regular expression beats a language model
Deterministic problems deserve deterministic solutions. A surprising amount of what gets handed to a model is a parsing problem with a known grammar.
035
Getting structured output from a model without tears
Asking for JSON and hoping is not a design. Schemas, validation and a bounded retry make the difference between a demo and something you can run unattended.
034
A migration wave plan that survives the change board
Sequencing is where platform migrations succeed or fail. Here is the ordering logic I use, and why the first wave should be boring on purpose.
033
What a GPU node costs before it produces a single token
The accelerators are the headline. Power, cooling, fabric, storage and the eighteen weeks of waiting are the budget.
032
What to do with the VMs nobody can identify
Every estate has them. Powered on, consuming capacity, owned by no one, named something like SRV-TEMP-02. Here is a decommissioning process that is safe and actually completes.
031
Hyper-V in 2026: the option people keep forgetting to price
It is in the licence you already own, it runs production for very large organisations, and it is routinely left off shortlists for reasons that stopped being true a decade ago.
030
Proxmox in the enterprise: where it fits and where it does not
It went from a homelab favourite to a serious contender fast. The technology is sound; the questions worth asking are about support, scale and who operates it.
029
Identity is the blast radius
Every control you have built sits behind a credential. Map which credentials reach which systems and the security architecture usually looks worse than the diagram suggests.
028
Segmentation without buying a microsegmentation product
Most of the value is available from equipment you already own. The hard part was never the enforcement technology, it is knowing what talks to what.
027
Certificate expiry is an availability risk, not a security task
A certificate that expires takes the service down as effectively as a failed disk. It is filed under security, owned by nobody, and tracked in a spreadsheet.
026
Ansible for infrastructure teams who have no automation yet
Starting from a pile of hand-built servers. The first thing to automate is not provisioning, it is the thing you do every month and get slightly wrong.
025
Idempotency is the whole idea, and most playbooks do not have it
A task that is safe to run twice is a different kind of object from a script. Getting this right is what separates configuration management from remote scripting.
024
Inventory and secrets: the two things that decide whether automation scales
A static host file and a password in a variable will work for twenty machines and become the reason the project stops at two hundred.
023
Terraform and Ansible are not competitors
One creates things that did not exist. The other configures things that do. The confusion comes from both being called infrastructure as code.
022
Infrastructure as code on an estate you did not build
Every tutorial starts from nothing. Real environments start from fifteen years of accumulated configuration, and the strategy for getting there is different.
021
Testing infrastructure code, without building a software team
Your playbooks now change production. That makes them software, and the minimum viable testing discipline is smaller than you fear.
020
Open networking: the switch and its operating system are now two purchases
Disaggregation did to networking what it did to servers twenty years earlier. The technology works; the interesting part is what it does to your operating model.
019
Nobody sizes the fabric, then blames the GPUs
East-west traffic in an AI cluster behaves nothing like the enterprise traffic your network was designed for. The symptom looks like slow accelerators.
018
SD-WAN: what it actually solves, and what it quietly moves
The pitch is cheaper circuits. The real value is that the WAN becomes centrally managed and application-aware. The cost is a new dependency you should look at closely.
017
5G and NFV infrastructure: telco workloads on general-purpose kit
The network function became software running on servers you recognise. The requirements it brings with it are not the ones enterprise virtualization taught you.
016
Seismic interpretation belongs in the data centre, not under a desk
Oil and gas workstations are the strongest remaining case for GPU-backed virtual desktops. The data is already in the data centre and the workstation keeps travelling to meet it.
015
How much accelerator memory a model actually needs
Parameters times bytes per parameter is the start. What catches people is everything else in the memory budget, and it is not small.
014
The GPU utilisation metric is lying to you
A device reporting 100% utilisation may be doing almost nothing useful. The number means something narrower than its name suggests.
013
Sharing accelerators: time-slicing, partitioning, and what each one costs
One device, several consumers. Three mechanisms with genuinely different isolation properties, and choosing by convenience produces surprises.
012
Air, rear door, or direct liquid: the cooling decision is a building decision
Rack density crossed the point where air alone stops working. Choosing what replaces it commits your facility for a decade, and IT is not the deciding party.
011
Sovereign AI infrastructure, minus the press release
The phrase covers four different requirements with four different technical answers. Deciding which one you actually have is most of the work.
010
Network automation: start by reading, not by writing
The fear that stops network teams automating is a bad push taking the network down. The first six months should not involve pushing anything.
009
Patching infrastructure that cannot reboot
The cadence is set by the reboot, not by the patch. Systems with no maintenance window accumulate risk until something forces the issue.
008
The credentials nobody rotates
Service accounts, API keys, and the shared password in a document from 2019. They do not expire, nobody owns them, and they are what an intruder looks for first.
007
Logging enough to investigate, not enough to bankrupt you
Log everything and the bill becomes the constraint. Log the wrong things and the investigation fails. The selection is the skill.
006
Capacity forecasting without a crystal ball
Most capacity plans are a straight line drawn through two points. Three cheap improvements make them substantially more useful than that.
005
Indexing 604 pages of recitation with Whisper and a lot of fuzzy matching
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.
004
RAG is a retrieval problem wearing an AI costume
Nine out of ten disappointing RAG systems are disappointing search systems. The generation step is rarely where the failure is.
003
NVMe/TCP quietly won the mid-range while everyone argued about Fibre Channel
Not because it is faster. Because it removed a team, a skill set and a second network from the build sheet.
002
Your immutable backup is not an air gap
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.
001
The three questions that end most 'we will just move to the cloud' plans
Not arguments against public cloud. Just the three numbers that tend to be missing when the business case looks obviously good.