Network Throughput Calculator for a NAS

Two numbers decide how fast a file moves between a computer and a NAS: what the link can carry and what the array can supply. The slower of the two is your transfer speed, and on a home network it is almost always the link below 2.5 gigabit and almost always the drives above it. Knowing which side you are on decides whether the next hundred dollars belongs in a switch or in drives, which is the question this calculator answers.

Why is my NAS slower than the speed on the box?

Network links are sold in bits per second and file transfers are measured in bytes per second, and there are eight bits in a byte. That single conversion accounts for most of the gap between what people expect and what they see. A gigabit link is 1,000 megabits per second, which is 125 megabytes per second before any overhead at all.

Then subtract the overhead. Ethernet framing, IP and TCP headers, acknowledgements and the file-sharing protocol on top consume roughly 5 to 10 percent of the wire. The practical ceiling on gigabit is about 113 MB per second, and a transfer sitting at 110 is not slow, it is the wire completely full. No amount of tuning recovers the missing 890, because the missing 890 never existed.

Hours per TB is the time to move one decimal terabyte at the realistic rate, on large files, with nothing else competing for the link. Wi-Fi figures assume a strong signal on a clean channel and a single client, which is the best case rather than the usual one.

Is the network or the array the bottleneck?

A single 7200 rpm 3.5 inch drive sustains 150 to 260 MB per second on sequential reads. Read that against the table above and the shape of the problem becomes clear immediately: one drive already outruns gigabit by a comfortable margin, so on a 1GbE network every multi-bay NAS in existence is waiting on the network.

Move up and the crossover arrives quickly. Four mechanical drives in RAID 5 or RAIDZ1 sustain roughly 450 to 700 MB per second sequentially, which fits inside 5GbE and leaves a 10 gigabit link half empty. Filling 10GbE with mechanical drives takes eight or more spindles in a wide stripe, or a flash pool.

The practical rule for a home server: 2.5GbE is the upgrade that pays, 10GbE usually is not. Going from 1GbE to 2.5GbE roughly triples real throughput for the price of a small switch and a couple of adapters, and every mechanical array can supply it. Going from 2.5GbE to 10GbE costs several times more and only pays back if the pool is flash or very wide. The same money spent on drives buys capacity you will definitely use.

Two caveats keep that rule honest. Random access, which is what virtual machine images and databases produce, is nothing like sequential and mechanical drives collapse under it. And a NAS serving several people at once splits its array throughput between them, which is the one home case where aggregation and a faster link genuinely help.

Why do small files transfer so much slower?

Every file costs a fixed set of operations no matter how small it is: open, query metadata, write, set attributes, close. At 100,000 files those round trips dominate completely and bandwidth stops being the limit, which is why a folder of photos can move at a small fraction of the rate a single video file reaches over the same wire.

This is the effect the file-profile dropdown in the calculator models. It is also why the fix is not a faster network. Three things that do help:

• Archive before transferring. Rolling a directory of small files into a single container turns hundreds of thousands of operations into one large sequential stream.
• Put small-file workloads on flash. A SATA SSD handles the operations per second that a mechanical drive cannot, and it matters far more here than the link speed does.
• Use a protocol suited to the job. Bulk sync tools that batch operations outperform a drag-and-drop copy on the same hardware, sometimes by a large factor.

What does it cost to move from gigabit to 2.5GbE?

Less than most people expect, which is what makes it the best value change in a home lab. You need three things: a switch with multi-gigabit ports, a port on the NAS, and a port on the client that actually matters.

Prices are the listing prices recorded at verification and move often. Cabling is usually not a cost: existing Cat5e runs carry 2.5GbE reliably at household lengths, and Cat6 covers 10GbE at the distances found in a house.

Notice what is missing from that table, which is new cable. 2.5GbE was designed specifically to run over the Cat5e already in the walls, which is the reason it exists and the reason it is the cheap upgrade. 10GbE over copper is stricter and wants Cat6 or better, though at the distances inside a house Cat6 is entirely sufficient and Cat6a is rarely needed.

Does link aggregation help a single transfer?

No, and this is the most common disappointment in home networking. Standard link aggregation places each connection on one physical port by hashing its addresses, so a single file copy between two machines rides one link and reaches one link of throughput. Two bonded gigabit ports do not produce a two gigabit transfer.

What aggregation does buy is total throughput across several simultaneous conversations, plus failover if a cable or a port dies. On a NAS serving four people at once that is a real benefit. For the person watching one progress bar it is none at all, and the money is better spent on a single faster link.

How do I find out what my network is really doing?

Measure the link and the disk separately, because a single file-copy number cannot tell you which one is the limit.

• Test the link with a memory-to-memory tool such as iperf3, which moves generated data and never touches a disk. If that reports close to the realistic figure in the table above, the network is healthy and any slower file copy is a storage or protocol problem.
• Test the array locally on the NAS itself, writing and reading a large file to the pool. That gives you the array ceiling with the network removed entirely.
• Then compare. If the link test is fast and the local disk test is fast but the file copy is slow, the answer is usually small files, encryption on an underpowered processor, or a client-side limit.

One frequently missed culprit is the client. A laptop on Wi-Fi, an old USB enclosure, or a machine whose own drive cannot write faster than 120 MB per second will cap the transfer no matter what the server and the switch can do. That is what the client ceiling field in the calculator is for.

Related reading

• NAS networking basics for the full explanation behind these numbers
• 2.5GbE against 10GbE for the upgrade decision in detail
• Best 2.5GbE switches for a NAS for the picks
• Hard drives against SSDs for a NAS for whether flash fixes your bottleneck
• RAID capacity calculator because layout changes array throughput as well as capacity