Quick Summary (TL;DR): A NASPriceAliExpress price stuck around 11 MB/s often has a 100Mb/s link; around 110 MB/s can be normal for a healthy 1GbE path. A 2.5GbE connection has a theoretical ceiling of 312.5 MB/s, while 10GbE has a theoretical ceiling of 1,250 MB/s—but real file copies also depend on disks, RAID, CPU, SMB, client storage and the slowest link in the route. Do not buy a 10GbE switchPriceAliExpress price or SSD cache until you have measured the network independently of the storage. Check negotiated speeds at both ends, run iperf3 in both directions, then compare a large-file copy with NASAliExpress price disk and CPU activity. This guide gives a repeatable, safe diagnosis and explains what each result means.
Scope: This is a troubleshooting guide for local NAS file transfers over Ethernet, principally SMB, not a hardware benchmark or a claim that any particular NAS model reaches a given speed. The commands are examples to run only on systems you administer; interface names, package availability and vendor menus vary.
The First Question: What Speed Should Your NAS Actually Reach?
Network cards advertise gigabits per second (Gb/s); file managers commonly show megabytes per second (MB/s). Eight bits make one byte. Ethernet, IP, TCP and SMB consume some of the line rate, so the headline link speed is a ceiling, not a promised file-copy result. An SSD or RAID array cannot make a 1GbE link carry 500 MB/s to one computer.
| Negotiated link | Theoretical decimal payload ceiling | Indicative healthy large-file transfer* | What to investigate first |
|---|---|---|---|
| 100MbE | 12.5 MB/s | About 9–12 MB/s | Unexpected 100Mb/s negotiation, cable or port |
| 1GbE | 125 MB/s | Often roughly 100–118 MB/s | Check whether this is simply the link limit |
| 2.5GbE | 312.5 MB/s | Often roughly 230–290 MB/s | Check both ends and switch uplinks |
| 5GbE | 625 MB/s | Highly dependent on endpoints | Check CPU, adapters and storage |
| 10GbE | 1,250 MB/s | Hundreds of MB/s to above 1 GB/s with suitable storage | Measure disks and CPU as well as the network |
*Illustrative planning ranges, not measured results or guarantees. They assume a stable wired local connection, a large file and sufficiently fast client/server storage; protocol, hardware and workload differences can move results substantially. 1 GB/s means 1,000 MB/s, whereas 1 GiB/s is approximately 1,074 MB/s.
A reported “gigabit” transfer of 110 MB/s is not a malfunction. Equally, a 10GbE NICPrice displaying “10 Gbps connected” does not prove the NAS can read or write at 1 GB/s. Check the whole path: client NIC → patch cable → switch port → switch uplink (if any) → NAS NIC → CPU → storage pool.
A Symptom-to-Cause Map
| Observed symptom | Most likely class of bottleneck | Next measurement |
|---|---|---|
| About 10–12 MB/s on a wired LAN | 100Mb/s negotiation or a 100Mb/s hop | Inspect every Ethernet link and switch port |
| About 100–115 MB/s, very stable | 1GbE ceiling | Confirm link speeds; use iperf3 |
| Around 200–280 MB/s on 2.5GbE | May be healthy, or storage-limited | Compare iperf3 and large-file copy |
| 10GbE link but file copy near 110 MB/s | A 1GbE hop, selected NIC, or disk bottleneck | Check path and SMB interface selection |
| Fast first seconds, then much slower | Write cache exhaustion, sustained HDD write speed, SMR or thermal limits | Long copy plus disk and temperature monitoring |
| Large videos fast; folders of photos slow | Metadata, latency, antivirus/indexing, many tiny files | Compare one large file with many small files |
| Reads fast, writes slow | Write-side disk/RAID/SMB or client-source issue | Reverse iperf3 and compare NAS write load |
| Only Wi-Fi devices are slow | Wireless link quality, airtime or backhaul | Repeat using wired Ethernet |
| Only remote/VPN transfers slow | Internet upload, VPN CPU, latency or relay | Separate LAN test from WAN test |
Use This Diagnostic Order, Not Random Tweaks
- Record the problem: source and destination, file count/size, direction, protocol, observed sustained MB/s and whether the client is wired.
- Check negotiated Ethernet speeds: both client and NAS, every switch port and any intermediate uplink.
- Run iperf3: forward, reverse and optionally parallel TCP streams, ideally with no disks involved.
- Measure storage: copy a genuinely large file while watching disk utilisation, CPU, network graphs and background tasks.
- Investigate the application layer: SMB configuration, client security scanning, permissions, snapshots, encryption and small-file overhead.
- Change one thing: retest with the same file and route; buy hardware only after identifying the actual ceiling.
Synology’s own slow-transfer guidance separates client, network and NAS causes and recommends sufficiently large test files; its troubleshooting article cites at least 5 GB for useful copy testing. The same separation applies to QNAP, UGREEN, TrueNAS and DIY NAS systems. Source: Synology Knowledge Center.
Start with a Reproducible Baseline
Before changing settings, note the NAS model and OS version, number and type of drives, RAID or ZFS layout, whether the pool is rebuilding or scrubbing, and the client’s operating system. Record the switch model, ports in use and whether a dock or USB Ethernet adaptor is involved. Disconnect VPN software from the local transfer path only if doing so is safe and authorised. Avoid testing during backup jobs, parity checks or media indexing unless those tasks are precisely what you want to investigate.
| Record | Example of useful detail | Why it matters |
|---|---|---|
| Client and NAS | Windows PC → NAS SMB share | Direction determines read/write load |
| Route | PC 2.5GbE → switch 2.5GbE → NAS 2.5GbE | A single 1GbE uplink caps the path |
| Dataset | One 8GB video file vs 30,000 photos | Large-file throughput differs from metadata-heavy copying |
| Time and speed | After 30 seconds: sustained MB/s | Short peaks can reflect cache |
| CPU, RAM, disks | Per-core CPU; pool activity; temperature | Reveals resource constraints |
| Background work | RAID rebuild, scrub, indexing, backup | Competing I/O can distort results |
Keep an untouched copy of your test data. Use a dedicated temporary test folder with enough free space; remove only files you created for the test. Never benchmark a production disk with destructive raw-device write commands.
Check That Every Ethernet Link Negotiated the Expected Speed
A 2.5GbE NAS connected to a 2.5GbE PC is still limited to 1GbE if the path traverses a 1GbE switch uplink. A 10GbE switch cannot accelerate a NAS with only a 1GbE NIC. Likewise, a damaged cable, poor termination or unsuitable adaptor can cause a link to negotiate at 100Mb/s or flap intermittently. Read the actual negotiated speed, not just the model’s advertised capability.
Windows: Check Link Speed and the Active Adaptor
|
1 2 3 4 |
Get-NetAdapter | Format-Table Name, Status, LinkSpeed, InterfaceDescription -AutoSize Get-NetIPConfiguration |
Run in PowerShell. Look for the adaptor carrying the NAS traffic: Ethernet, a USB NIC, a dock NIC or Wi-Fi. If the active interface shows 1 Gbps, a 10GbE NAS on the far side will not change that client’s single-link ceiling. If both Wi-Fi and Ethernet are connected, confirm which IP route is used. Microsoft documents Get-NetAdapter as the command for viewing basic adaptor properties.
Linux: Inspect the NIC and Error Counters
|
1 2 3 4 5 6 7 |
ip -br link ip route get 192.0.2.20 sudo ethtool enp3s0 ip -s link show dev enp3s0 sudo ethtool -S enp3s0 |
Replace 192.0.2.20 with the NAS address and enp3s0 with your actual interface. The documentation-only 192.0.2.0/24 address is not a real NAS address. ethtool reports negotiated speed and duplex where the driver supports it; error counters can reveal dropped packets, physical-layer faults or driver problems. A counter changing under load is more useful than a large historical total. Counter names are driver-specific.
macOS, NAS Interfaces and Switches
On macOS, inspect the Ethernet service in System Settings → Network and the adaptor’s hardware details; menus differ by macOS release and adaptor. On Synology DSM inspect Network Interface under Control Panel → Network. On QNAP QTS/QuTS hero use Network & Virtual Switch → Interfaces. For other NAS brands, use the vendor’s network interface page. Managed switches usually expose per-port link rate and CRC/FCS counters. For an unmanaged switch, check its documented port capability and LEDs, then test by substitution if necessary.
If you see 100Mb/s where 1GbE or 2.5GbE is expected, first replace the patch cable with a known-good, correctly terminated cable and try a different switch port. Inspect wall sockets and keystone terminations; Gigabit copper Ethernet requires all four twisted pairs. Leave autonegotiation enabled unless the equipment vendor specifically instructs otherwise. Record the original configuration before altering anything.
Run iperf3 to Separate Network Throughput from Disk Throughput
iperf3 measures network traffic between two endpoints without the normal NAS file-share read/write path. That makes it the most useful next test. ESnet’s official documentation explains client/server operation, reverse mode (-R), parallel streams (-P) and the default TCP port 5201. Install a supported build on a client and a NAS-compatible host, container or temporary Linux machine. Some NAS appliances do not include iperf3 and may restrict shell/package installation; do not bypass vendor support or alter a production appliance simply to run a benchmark.
If running iperf3 directly on the NAS is unsupported, put a second capable wired computer on the NAS switch port or a known equivalent port. That tests the network path, though not the NAS NIC or NAS CPU itself; be explicit about what the substitute test does and does not prove.
A Safe Two-End iperf3 Test
On the receiving test host (for example, a supported Linux box on your LAN):
|
1 2 3 |
iperf3 -s |
From the other host, replace 192.0.2.20 with the test server’s actual local IP:
|
1 2 3 4 5 6 |
iperf3 -c 192.0.2.20 -t 20 iperf3 -c 192.0.2.20 -t 20 -R iperf3 -c 192.0.2.20 -t 20 -P 4 iperf3 -c 192.0.2.20 -t 20 -P 4 -R |
The first command sends client → server; -R reverses the data direction; -P 4 opens four parallel TCP streams. Compare the sender and receiver summaries and note retransmissions. Run each test more than once. An iperf3 result is reported in bits per second; divide by eight to compare approximately with MB/s, remembering protocol overhead and decimal/binary units.
For example, 2.3 Gbit/s ÷ 8 = 287.5 MB/s is a unit conversion, not a claim about any specific NAS. A good iperf3 result with poor SMB file-copy results points you towards storage, CPU, protocol or application behaviour. Poor iperf3 and poor SMB together point towards the network path or endpoint networking resources.
Security: run the iperf3 server only on a trusted local network, restrict TCP port 5201 in your firewall, and stop the server after testing. Do not forward this port from the internet or expose a diagnostic container to an untrusted network.
Interpret One Stream Versus Four Streams Carefully
| iperf3 outcome | Likely interpretation | Follow-up |
|---|---|---|
| One and four streams both near expected line rate | Network path probably healthy | Test actual storage and SMB |
| One stream slow; four much faster | Per-flow CPU, TCP processing, RSS or host limits may matter | Check per-core CPU and NIC features |
| Both directions slow | Link, switch, NIC, CPU or test host bottleneck | Check negotiation, errors and endpoint capability |
| Only reverse direction slow | Asymmetric receive/send or endpoint limitation | Inspect receiver counters and CPU |
| Retransmissions rise under load | Possible loss, congestion, driver or physical issue | Inspect switch CRC/FCS and NIC errors |
| Fast between PCs, slow when NAS is an endpoint | NAS NIC, NAS CPU, driver or virtualisation path | Test NAS networking and resource usage |
Parallel streams do not magically make a single SMB file copy use four connections. Their purpose here is to distinguish a per-flow or per-core limit from the available network capacity. iperf3 itself consumes CPU; a low-power NAS or test host can become the limiting factor, particularly on 10GbE. From iperf3 3.16, the upstream project documents separate threads for parallel streams, so software version can affect how a multi-stream test behaves.
Test the NAS Disks Separately from the Network
Once the network has been measured, assess the storage path. A NAS with one mechanical HDD cannot be assumed to sustain 10GbE writes. RAID, ZFS record size, file fragmentation, snapshots, encryption, drive model, thermal state and pool fullness can all change the result. Conversely, an SSD pool capable of much higher speeds can still appear “slow” over 1GbE.
- Copy one large, already-existing, incompressible file (ideally 5–20GB) from client to NAS and back, timing each direction. Use enough data to exceed short-lived caches where practical.
- Watch NAS CPU utilisation per core, not only the average; a single saturated thread can cap a transfer while overall CPU looks modest.
- Watch disk busy time, I/O wait, pool throughput, device latency and network traffic simultaneously. On ZFS,
zpool iostatis useful if shell access is supported. - Repeat with a folder of many small files. A large-file test that is fast and a small-file test that is slow usually calls for a different solution.
- Check pool health, SMART warnings, degraded RAID state, scrubs, rebuilds, snapshots, antivirus/indexing and active backup tasks.
Read-Only Monitoring Commands on a Linux or ZFS NAS
|
1 2 3 4 5 |
iostat -xz 1 zpool iostat -v 1 top |
Run only commands available and supported on your platform. iostat is normally supplied by the sysstat package; zpool iostat applies to OpenZFS, not every NAS. These commands observe activity rather than writing benchmark data. Use the NAS dashboard if shell access is unavailable. The Linux kernel documents the underlying block-device I/O statistics; TrueNAS also advises testing disk and network separately and warns against destructive write/trim benchmarks on active storage.
For a client-side sanity check, compare the source and destination disks with an ordinary local copy, or a trusted storage benchmark using a dedicated test file. Do not benchmark directly against raw disks or active pool devices. Be aware that RAM cache can inflate short read tests, and a thin or sparse test file can give misleading results.
Why a Transfer Starts at 900 MB/s and Falls to 150 MB/s
A high initial number can reflect RAM buffering, a fast SSD write cache, or a file manager’s progress estimation. Once dirty data must be committed to slower HDDs, the long-run speed falls. The correct question is not “what peak did I see?” but “what throughput was sustained after several gigabytes, and what was the NAS doing when it changed?”
This pattern can also occur when a drive becomes hot and throttles, a pool approaches full capacity, a rebuild competes for I/O, or a device-managed SMR HDD suffers long sustained writes. Do not diagnose SMR solely from a graph; verify the exact drive model, firmware and workload. A NAS cache upgrade is justified only after showing that the workload is cache-friendly and that the appliance supports the intended cache or storage-pool configuration.
For the buying decision, see NAS NVMe Cache vs SSD Storage Pool: Which Should You Use?. It explains why an SSD storage pool and an SSD cache solve different problems.
SMB Can Be the Bottleneck Even When iperf3 Is Fast
SMB adds file-system semantics, authentication, locking and metadata work that iperf3 deliberately avoids. SMB signing or encryption, CPU limits, antivirus, snapshot handling and a client’s storage performance may matter. Do not disable security controls just to improve a benchmark; first identify whether a feature is required and whether the NAS CPU or protocol stack is actually saturated.
Windows SMB Diagnostics
|
1 2 3 4 5 6 |
Get-SmbConnection Get-SmbClientConfiguration | Select-Object EnableMultichannel Get-SmbClientNetworkInterface Get-SmbMultichannelConnection -IncludeNotSelected |
Run while a large SMB transfer is active. Get-SmbConnection identifies the negotiated SMB dialect and connections. The other commands help reveal whether the Windows client sees suitable interfaces and whether SMB Multichannel has selected them. Microsoft documents these exact checks for SMB Multichannel. NAS-side SMB Multichannel support is vendor-, model- and firmware-dependent; enabling a setting on Windows alone does not guarantee that the NAS can use it.
If the copy uses the wrong interface, inspect routing, IP addressing and NAS binding rather than assuming that the fastest port is automatically selected. If the Windows client reports multiple paths, check that they actually carry traffic. Record SMB version and security policy; do not force obsolete SMB1 or weaken signing to chase a headline number.
Two 1GbE Ports Do Not Automatically Give One PC 2Gb/s
Link aggregation (LACP/802.3ad) usually distributes separate traffic flows across links; a single conventional TCP flow generally remains limited to one physical member. LACP is useful for multiple simultaneous clients and resilience when correctly configured, but is not a substitute for a genuine 2.5GbE or 10GbE path for one fast workstation. SMB Multichannel can use multiple connections where both ends and the network design support it. It is a separate feature, not a blanket promise that any two-port NAS will double single-file speed.
For a focused comparison of network upgrade value, see 2.5GbE vs 10GbE at Home: Is the Upgrade Worth It?.
Many Small Files Are a Different Workload
A folder containing 100,000 photographs, source files or tiny documents creates many file opens, metadata lookups, directory updates and sometimes permission checks. Each operation has latency. The displayed MB/s can be low even when neither the network nor the disks are near their sequential throughput limits. Wi-Fi latency and a slow HDD’s random I/O make this more noticeable.
Compare a single large file of similar total size with the small-file folder. If the large file is fast, focus on metadata latency, source and destination IOPS, client security scanning, NAS indexing and protocol overhead. For an occasional migration, packaging small files into an archive before transfer can reduce per-file overhead, provided you verify integrity and unpack safely at the destination. Do not archive live application databases or containers without a consistent backup procedure.
Check Cables, Switches, Transceivers and Packet Errors
The physical path deserves attention even if the connection is “up”. Copper links can suffer bad patch leads, damaged wall cabling, marginal connectors or intermittent autonegotiation. At 10GbE, SFP+ DAC, optical transceivers and 10GBASE-T modules must be supported by the particular NIC and switch; module power, heat and distance ratings also matter. A switch that advertises 10GbE ports may have a slower uplink or a limited mixed-speed topology.
| Observation | Physical/network check | Practical response |
|---|---|---|
| 100Mb/s negotiated | Patch cable, wall socket, switch port | Swap one cable/port at a time |
| 2.5GbE endpoints but 1GbE throughput | Intermediate 1GbE port or uplink | Map each hop; move to multigig port |
| Link flaps under load | Cable, module, thermal or power problem | Inspect logs, temperatures, counters |
| CRC/FCS errors increase | Physical-layer integrity | Test cable/connector/transceiver |
| 10GbE RJ45 module runs very hot | Module and switch power/thermal budget | Use supported modules and adequate airflow |
| Direct connection faster than through switch | Switch configuration, cabling or uplink | Isolate suspect switch path |
Do not assume Cat5e is automatically unsuitable for 2.5GbE; the standard was designed to work with existing twisted-pair installations under specified conditions. For new long 10GBASE-T runs, choose suitable certified cabling and installation practices. Where both devices offer SFP+, a short supported DAC may avoid some RJ45 transceiver power and heat, but compatibility must be checked.
For network hardware options and their trade-offs, read Best 10GbE Switches for NAS and Home Servers and SFP+ DAC vs 10GBASE-T: Cables, Heat, Distance and Total Cost.
Jumbo Frames Are Not the First Fix
The standard Ethernet IP MTU is typically 1500 bytes. Jumbo frames (often an MTU of 9000) reduce packet processing overhead in some high-throughput workloads, but require consistent support along the relevant path. Mismatched MTUs can cause hard-to-diagnose connectivity and performance problems. QNAP’s documentation explicitly notes that all connected devices in a jumbo-frame path must support the configured MTU and that support varies by model.
Start and benchmark at MTU 1500. Consider jumbo frames only after the network is otherwise healthy, all participating endpoints and switches support the chosen value, and you can safely test and roll back. A move from 1500 to 9000 will not cure a 1GbE uplink, slow disk or failing cable. On routed networks, VPNs and mixed-client environments, the operational cost can outweigh any small benefit.
Wi-Fi and Remote Access Need Separate Expectations
Wi-Fi link rates are not equivalent to sustained file throughput. Airtime sharing, signal strength, interference, channel width, access-point backhaul and roaming all affect results. For a reliable NAS benchmark, temporarily use a wired client and compare. If wired transfers are fast and Wi-Fi is slow, investigate the wireless environment rather than buying NAS disks.
Remote transfers have a different ceiling: the slower site’s internet upload speed, VPN processing, latency and possible relay paths. An excellent 10GbE LAN will not make a remote upload exceed a modest broadband upstream rate. Compare local SMB performance before testing a VPN or remote-access app. For secure access architecture, see Remote NAS Access: Tailscale, WireGuard or the Manufacturer’s App?.
Four Worked Diagnostic Scenarios
Scenario: 11 MB/s on a “Gigabit” Network
A home user copies a large file at approximately 11 MB/s. This is close to the 100Mb/s line-rate ceiling of 12.5 MB/s. The first check is negotiated link speed at the PC, NAS and switch, not SSD caching. If one port reports 100Mb/s, swap the patch lead and check any wall termination or intermediate switch. Only after all links negotiate correctly should you test storage.
Scenario: 110 MB/s on 1GbE
A healthy 1GbE path often tops out near 110 MB/s for a large SMB file. If iperf3 is near line rate and a large file copies at this level, the network is behaving plausibly. Faster disks will not help this particular one-client transfer until the network is upgraded. A 2.5GbE NIC and switch path may be a more cost-effective next step than 10GbE if the storage can sustain it.
Scenario: 10GbE Negotiates Correctly but SMB Reaches Only 180 MB/s
Suppose iperf3 reports several gigabits per second in both directions, but a long NAS write settles near 180 MB/s. This hypothetical pattern makes the storage path or SMB implementation the next suspect. Inspect disk activity, pool health, CPU per core and any write cache; compare read versus write. Do not buy another 10GbE switch merely because the copy is slow.
Scenario: One Video Fast, Thousands of Photos Slow
If a large video file copies quickly but many small photos crawl, the path has already demonstrated that it can carry large sequential data. Focus on directory metadata, IOPS, client-side scanning, indexing and filesystem latency. SSD storage for metadata-heavy applications may help more than extra network bandwidth, but verify that the workload and NAS software support the proposed change.
A Decision Table: What Is Worth Buying?
| Measured finding | Usually sensible next step | Do not buy yet |
|---|---|---|
| 100Mb/s link or rising CRC errors | Correct cable/termination or port fault | New NAS, SSD cache or 10GbE switch |
| Healthy 1GbE ceiling and faster local disks | 2.5GbE NIC + suitable switch/path | 10GbE without a workload case |
| 2.5GbE already saturated; fast SSD or multi-disk pool | Evaluate 10GbE NIC/switch and cables | Extra HDDs solely for network speed |
| iperf3 fast, sustained HDD writes slow | Inspect pool, drives, rebuilds and write workload | More Ethernet bandwidth |
| Small-file IOPS or metadata bottleneck | Consider SSD pool where supported | Network upgrades based on small-file MB/s |
| One CPU core saturated during SMB | Check platform CPU and supported SMB features | More switch ports |
| Slow only over Wi-Fi | Improve AP/backhaul and client radio path | NAS drive upgrades |
| Slow only over VPN/WAN | Check internet upstream and VPN route/CPU | 10GbE LAN upgrade for remote speed |
If you do need a switch, compare actual port speeds, uplink topology, SFP+ versus RJ45, heat, noise and power draw—not just the number of ports. The relevant purchase guide is Best 10GbE Switches for NAS and Home Servers. For an affordable intermediate step, Best 2.5GbE Switches for a Smart Home and Home Lab covers multi-gigabit selection.
A Reusable NAS Transfer Test Sheet
| Test | Result to record | What it isolates |
|---|---|---|
| PC and NAS link speed | Negotiated Gb/s and active interface | Hard link-rate ceiling |
| Switch ports and uplinks | Every hop’s negotiated rate | Hidden 1GbE or 100Mb/s bottleneck |
| iperf3 client → server | Gbit/s; retransmissions | Network send direction |
| iperf3 reverse | Gbit/s; retransmissions | Network receive direction |
| iperf3 with four streams | Gbit/s; CPU per core | Per-flow versus aggregate limit |
| Single 5–20GB file to NAS | Sustained MB/s after initial cache | NAS write path plus SMB |
| Same file from NAS | Sustained MB/s | NAS read path plus SMB |
| Thousands of small files | File count, total size, elapsed time | Metadata and IOPS behaviour |
| NAS resource dashboard | CPU, disks, pool, temperature, background jobs | Local resource bottleneck |
Repeat the same measurements after one change. Keep a short note of firmware versions and hardware revisions; the result is more useful than a screenshot of one peak transfer rate. If the NAS is a production backup target, schedule tests so they do not interfere with backup windows and verify data integrity afterwards.
Frequently Asked Questions
Why is my NAS limited to 100 MB/s?
The most common benign explanation is a 1GbE link: 1Gb/s equals 125 MB/s before overhead. A sustained copy near 100–115 MB/s can be normal. Check actual negotiated speed and use iperf3 before assuming the NAS is faulty.
Why does my 2.5GbE NAS transfer at only 110 MB/s?
Look for a 1GbE port or uplink anywhere between the client and NAS, including a dock, router LAN port or switch. Confirm both endpoints negotiate 2.5Gb/s and that traffic is using the expected interface. If the whole path is multigigabit, compare iperf3 with file-copy speed to identify a storage or SMB limit.
Why does 10GbE not give me 1,250 MB/s?
1,250 MB/s is a mathematical line-rate ceiling, not a file-copy target. Protocol overhead, NIC and CPU limits, SMB processing and the storage pool reduce achievable throughput. A single HDD, slow RAID write workload or busy NAS can be the limit even when the NIC reports 10Gb/s.
Will an NVMe cache make NAS transfers faster?
Only for workloads that benefit from the NAS’s supported caching implementation. A cache cannot exceed the Ethernet link rate and often does little for one-off sequential copies of large files. An SSD storage pool is a different design with different capacity and protection trade-offs.
Should I enable jumbo frames for 10GbE?
Not as a first step. Keep MTU 1500 while diagnosing. Consider larger MTUs only if the entire path supports them, the configuration is tested and you have a rollback plan. Jumbo frames cannot fix slow disks, poor cabling or an unexpectedly slow link.
Can two 1GbE ports replace a 2.5GbE port?
Not automatically. Conventional link aggregation helps distribute separate flows across members; one ordinary TCP flow normally stays on one member. SMB Multichannel can use multiple connections if both client and server support and select them, but it requires validation rather than assumption.
What if iperf3 is fast but SMB is slow?
The network may be healthy. Measure local source/destination storage, NAS CPU per core, pool activity, SMB dialect and selected interfaces, security scanning, encryption and file count. A single large-file copy versus many small files is an especially informative comparison.
The Bottom Line
Measure first, upgrade second. A 100Mb/s link, 1GbE ceiling, slow HDD write path and small-file metadata bottleneck can all look like “slow NAS transfers”, but require different fixes. Confirm every negotiated link, run iperf3 in both directions, and observe disks and CPU during a sustained large-file copy. Only then decide whether you need a better cable, a 2.5GbE switch, a 10GbE upgrade, SSD storage or simply a different workflow.
Related NAS and Home-Lab Guides
- 2.5GbE vs 10GbE at Home: Is the Upgrade Worth It?
- Best 10GbE Switches for NAS and Home Servers
- Best 2.5GbE Switches for a Smart Home and Home Lab
- NAS NVMe Cache vs SSD Storage Pool: Which Should You Use?
- SFP+ DAC vs 10GBASE-T: Cables, Heat, Distance and Total Cost
- Remote NAS Access: Tailscale, WireGuard or the Manufacturer’s App?
Datasheets and Official Technical Resources
- ESnet: iperf3 manual, test directions, streams and output
- Synology Knowledge Center: diagnosing low file-transfer rates (updated May 2025)
- Microsoft Learn: SMB Multichannel configuration and verification
- Microsoft Learn: Get-NetAdapter PowerShell reference
- QNAP QuTS hero: interface speed and jumbo-frame configuration
- TrueNAS: network optimisation and performance-testing guidance
- Linux kernel: ethtool network interface diagnostics
- Linux kernel: disk I/O statistics