SAS, SATA and NVMe
Three interfaces, three different purposes. The choice matters more than the headline capacity.
SAS
Serial Attached SCSI is the server interface. It is what we fit in most machines, and the reasons are unglamorous but real:
- Dual porting. A SAS drive can be reached by two controllers. That matters in redundant setups and it means a flaky path does not take the drive with it.
- Better error reporting. SAS tells you what went wrong in detail. SATA often just retries silently until something times out, by which point you have a stalled machine and no clue why.
- Deeper command queues, which is what keeps throughput up when many processes hit the disk at once.
- Duty cycle. SAS drives are specified for continuous operation. Plenty of SATA drives are specified for eight hours a day.
For a machine that runs flat out for years with nobody standing next to it, that combination is worth the price difference.
SATA
Cheaper per terabyte and perfectly adequate for bulk storage where the workload is sequential and the data is replicated somewhere else — backup targets, media archives, write-once storage. Single-ported, less informative when it misbehaves. We fit it when it is the right tool, not as a default.
NVMe
Flash connected straight to PCI Express, with a command set designed for flash rather than inherited from spinning disks. The difference is not marginal: a SATA SSD might deliver tens of thousands of random operations per second, an NVMe drive hundreds of thousands to over a million, at a fraction of the latency.
Ask for NVMe when your workload is small random reads and writes: database engines, busy mail spools, container registries, anything where the disk queue is the thing people complain about. Ask for it for the latency, not the sequential throughput figure on the box — that figure is the least useful number on the datasheet.
Choosing, by workload
- Database, heavy writes
- NVMe, mirrored. Enterprise models with power-loss protection, which is what makes a sudden power cut survivable rather than interesting.
- Web and application servers
- SAS SSD or NVMe. The working set is usually small; latency beats capacity.
- NVMe or SAS SSD. Mail is thousands of tiny files and metadata operations, which is the worst possible case for a spinning disk.
- Backups, archives, media
- SAS or SATA spinning disks, in bulk, in RAID. Sequential and cheap per terabyte.
- Mixed
- A pair of NVMe for the system and the hot data, spinning disks behind it for the bulk. Most real servers end up here.
Endurance, briefly
Flash wears out by writing. Drives are rated in drive writes per day over a warranty period — a 0.3 DWPD drive and a 3 DWPD drive can look identical on a spec sheet and have ten times the write life apart. If your application writes constantly, say so when you ask for a quote and we will fit accordingly. Watch it afterwards with SMART; wear-levelling count is the attribute that matters on flash.
Just tell us the workload
We build to order, so you do not have to pick a model number. Describe what the machine will do — how much data, how it is accessed, how badly a slow disk would hurt — and we will specify drives that suit it. That conversation takes five minutes and saves a migration later.
Still stuck? Mail support@novogara.com — an engineer answers, at any hour. Back to the knowledge base