NAS Networking Basics: Why Transfers Are Slower

The most common disappointment with a new NAS is that a gigabit connection moves files at about 113 MB per second rather than anything resembling the number on the box. Nothing is broken. Gigabit means 1,000 megabits per second, there are eight bits in a byte, and after Ethernet framing and TCP and the file sharing protocol take their share you are left with roughly 113 megabytes per second. That figure is a saturated link, not a slow one, and the only cure is a faster wire.

The second most common disappointment is discovering that a single hard drive already outruns gigabit. One 7200 rpm 3.5 inch drive sustains 150 to 260 MB/s sequentially, so 0.75 of one drive is enough to fill the link. Every drive you add to the array beyond the first contributes nothing to transfer speed until the network gets faster. That single fact reorders most home NAS upgrade plans.

What is the difference between megabits and megabytes?

Network speeds are quoted in bits per second and file sizes are quoted in bytes, and there are eight bits in a byte, so every network number has to be divided by eight before it means anything to a file copy. The convention is lower case b for bits and upper case B for bytes, which is a distinction of one keystroke carrying a factor of eight, and it causes more confusion than any other unit on a home network.

So 1 Gbps is 125 MB/s before overhead. 2.5 Gbps is 312.5 MB/s. 10 Gbps is 1,250 MB/s. Those are ceilings that the wire physically cannot exceed, and then you lose more. Ethernet wraps every frame in a header, a trailer and an inter-frame gap. TCP adds its own header per segment and spends bandwidth on acknowledgements. SMB or NFS adds a further layer of request and response. The net result is that a healthy link delivers roughly 88 to 90 percent of its theoretical byte rate for a large sequential file, and much less for anything else.

This is also where the other capacity confusion sits. Networking uses decimal prefixes throughout, while operating systems report file sizes in binary units, so a "100 GB" folder that Windows shows as 100 GB is actually about 107 decimal gigabytes on the wire. The same 9.05 percent gap that makes a 20 TB drive report as 18.19 TiB is explained in the TB against TiB chart.

What is the real throughput ceiling of each link speed?

Theoretical is the link rate divided by eight. Realistic is what a large sequential file transfer achieves over a healthy wired link with a client and a NAS both able to feed it. The Wi-Fi rows are best case for one nearby client with nothing else on the channel, and the efficiency column shows why they behave so differently from cable. Run your own file size and link speed through the network throughput calculator.

The interesting column is efficiency. Wired Ethernet delivers about 89 to 90 percent of its theoretical byte rate across every speed, which makes the numbers predictable: a 2.5GbE link is genuinely two and a half times a gigabit link. Wi-Fi delivers 37 to 62 percent of its advertised rate at best, because it is a shared, half-duplex medium where every device takes turns and every frame is acknowledged. That gap is the whole reason a NAS belongs on a cable.

Does one hard drive really saturate gigabit?

Yes, and this is the most useful thing on the page. A single modern 7200 rpm 3.5 inch NAS drive sustains 150 to 260 MB per second on sequential reads, depending on capacity and on whether the data sits on the fast outer tracks or the slow inner ones. Gigabit tops out at 113 MB/s.

Several consequences follow, and most of them save money:

• Adding drives does not make a gigabit NAS faster. A four-drive array has more capacity and more redundancy than a single drive, and exactly the same transfer speed over gigabit, because the wire was full before the second drive was installed.
• An SSD cache does nothing for sequential transfers on gigabit either. The cache accelerates random access and metadata, which is real, but the sequential number was never limited by the platters.
• 2.5GbE is the highest-value upgrade on most home networks, because it moves the ceiling to 280 MB/s, which is just past where a single drive tops out. A five-port 2.5GbE switch and a pair of USB 2.5GbE adapters is a small fraction of what a single NAS drive costs.
• 10GbE needs an array to feed it. At 1,100 MB/s you need roughly 5 drives striping together, or flash, before the wire is the limiting factor again. Buying 10GbE for a two-drive NAS moves the bottleneck onto the drives and buys very little.

The full argument about which of the two to buy, including the switch and cabling costs that dominate the decision, is in 2.5GbE against 10GbE.

Why do small file transfers never reach line rate?

Every file carries a fixed protocol cost regardless of its size: an open, a metadata exchange, a permissions check, the data itself, a close and an acknowledgement. On a 40 GB video that overhead is invisible. On a 20 KB text file it is most of the work, and the network spends its time waiting for round trips instead of moving bytes.

This is why copying a folder of documents feels catastrophically slower than copying a film of the same total size, and why the progress bar on a source code checkout crawls. Nothing is wrong. The limiting factor has moved from bandwidth to latency, and a faster link does not reduce latency.

Effective rates are representative of what a 280 MB/s link achieves at each file size profile, where the limiting factor shifts from bandwidth to per-file protocol round trips. The pattern holds at every link speed: the small file rows barely improve when you upgrade the wire.

The practical fix is to stop sending small files individually. Archive a directory of thousands of small files into a single container before moving it, and the transfer runs at sequential speed. Backup software that packs files into large chunks does this by design, which is why a backup job often outruns a manual drag and drop of the same data by a factor of ten.

Is SMB or NFS faster on a home network?

SMB is the file sharing protocol native to Windows and supported everywhere. NFS is the Unix equivalent, native to Linux and BSD. Both are mature, both saturate a gigabit link on sequential transfers, and the difference between them on a home network is much smaller than forum arguments suggest.

Modern SMB3 is genuinely fast, supports multiple channels, encryption and signing, and is the correct default for a mixed household because every operating system speaks it without configuration. NFS tends to edge ahead on many-small-file workloads because it is a lighter protocol with fewer round trips per operation, and it is the better choice for a Linux host mounting NAS storage for containers or virtual machine images.

Two things matter far more than the choice between them. First, disable SMBv1 everywhere, which is a security requirement rather than a performance one and is covered in hardening a NAS against ransomware. Second, if your NAS and client both support SMB multichannel and you have two links, it does what people incorrectly expect from link aggregation: it genuinely spreads a single transfer across multiple paths.

Does link aggregation make a single transfer faster?

No, and this is the most expensive misunderstanding in home networking. Link aggregation bonds two or more physical ports into one logical link, then assigns each individual connection to a physical port by hashing the source and destination addresses. A given conversation lives on one port for its entire life.

So two clients copying files at the same time can each get a full port. One client copying one large file gets one port and exactly the speed of one port. Bonding two gigabit ports gives you 113 MB/s for that copy, not 226. People buy a managed switch, spend an afternoon configuring the bond, and measure precisely the number they started with.

If you want one transfer to go faster, buy a faster single link. A TP-Link TL-SG108S-M2 at $59.99 gives every port 2.5GbE for less than the cost of a managed gigabit switch plus the afternoon. Aggregation is worth configuring only when many clients hit the NAS at once, which in a home is a media server with several simultaneous streams rather than one person moving a folder.

The 2.5GbE switch roundup covers which models are managed and which are not. For a plain speed upgrade you want unmanaged, which is cheaper and has nothing to configure. Managed switches such as the TP-Link TL-SG108E or the NETGEAR GS308E earn their keep for VLAN segmentation rather than for speed.

Do jumbo frames help?

A jumbo frame carries about 9,000 bytes of payload instead of the standard 1,500, so the processor handles roughly six times fewer frames for the same amount of data. That reduces interrupt load and header overhead, and on 10GbE it can be worth a genuine few percent plus a noticeable drop in processor usage.

On gigabit and 2.5GbE the benefit is small enough to be lost in measurement noise, and the risk is real. Every device on the path has to agree. The NAS, the switch, the client adapter and any intermediate switch all need the same setting. If one does not, oversized frames are silently dropped and the symptom is not slowness but stalls: transfers that start, hang, and time out. Diagnosing that costs more than the few percent was ever worth.

The sensible rule is to leave jumbo frames alone unless you are running 10GbE on a small set of devices you control completely, and even then to change one thing and measure before changing the next.

What cabling do you actually need?

Less than the marketing suggests. Cat5e carries 2.5GbE reliably at the run lengths found in a house, which is exactly why the 2.5GbE standard exists: it was designed so people could get past gigabit on cable already in the walls.

Distances are the standard maximum run lengths. Home runs are almost always under 30 m, which is why the higher categories rarely change anything. A pack of Cat6 patch cables is the sensible default for new short runs, because it costs almost nothing more than Cat5e and removes the question.

If your NAS has a spare PCIe slot rather than a multi-gigabit port, a 10Gtek 10Gb PCIe NIC at $99.99 is the cheap route to 10GbE on a DIY box. For a laptop or a sealed mini PC, a Plugable 2.5Gbps USB-C adapter gets the client side to 2.5GbE without opening anything. For SFP+ rather than copper, the MikroTik CRS305 is the usual small-scale answer.

Why is the client usually the bottleneck?

Because upgrading the NAS is the part people think about. A 2.5GbE NAS talking to a gigabit laptop through a 2.5GbE switch runs at 113 MB/s, because a chain runs at the speed of its slowest link and there are four links in it: the NAS port, the NAS side cable, the switch, and the client port.

Work through the whole path before spending anything:

• The client network port. Most laptops sold today still ship gigabit, or no Ethernet port at all. A USB adapter fixes it for the price of a takeaway.
• The client storage. Writing 280 MB/s into a laptop with a slow SATA SSD or, worse, a 5400 rpm laptop drive will not happen no matter what the network does.
• The switch. A gigabit switch in the middle of two 2.5GbE devices caps everything. Check what every port in the path negotiates, not just the ends.
• The router. If traffic crosses between subnets or VLANs, the router does the forwarding and many consumer routers are far slower at that than their switch ports are.
• The NAS processor. With encryption enabled on the share, a low-power NAS chip can become the limit at 10GbE speeds even though nothing else in the path is.

A quick sanity check: transfer a single large file and watch the rate. If it sits flat at 113 MB/s, something in the path is gigabit. If it sits flat at 280, you are on 2.5GbE and doing well. If it fluctuates wildly, look at the file sizes and at Wi-Fi before you look at anything else.

Is Wi-Fi ever good enough for a NAS?

For consuming, yes. For moving, no. Streaming a 4K film needs 25 to 80 Mbps, which is a small fraction of what any modern Wi-Fi link delivers, so a media library over Wi-Fi is entirely reasonable and always has been.

Bulk transfer is a different problem. Wi-Fi is half duplex and shared: only one device transmits at a time on a channel, every frame is acknowledged, and throughput falls with distance, with walls and with every other active device. The table above puts a good Wi-Fi 6 client at about 75 MB/s, which is 22.2 min to move 100 GB against 6 min on a 2.5GbE cable.

Never put the NAS itself on Wi-Fi. The server is the one device every client talks to, so putting it on the shared medium means every transfer crosses the air twice, once from the client to the access point and once from the access point to the NAS. That halves an already halved figure. Run a cable to the NAS even if every client is wireless.

If a cable to the NAS is genuinely impossible, a pair of powerline adapters or a wired backhaul to a mesh node beats Wi-Fi for the server leg. And if you are choosing between spending on Wi-Fi and spending on a wired upgrade, the 2.5GbE switch plus two adapters costs less than most access points and delivers a bigger improvement to the one workload you actually notice.

Related reading

• 2.5GbE against 10GbE for which upgrade is worth the switch cost
• Network throughput calculator for your own file size and link speed
• Best 2.5GbE switch for a NAS for managed against unmanaged
• Hardening a NAS against ransomware for SMBv1, VLANs and remote access
• NAS drive specs chart for sustained transfer rates by model