Raspberry Pi holds its own against low-cost ARM NAS

Earlier this year, I pitted the $549 ASUSTOR Lockerstor 4 NAS against a homebrew $350 Raspberry Pi CM4 NAS, and came to the (rather obvious) conclusion that the Lockerstor was better in almost every regard.

Well, ASUSTOR introduced a new lower-cost NAS, the $329 Drivestor 4 Pro (model AS3304T—pictured above), and sent me one to review against the Raspberry Pi, since it make for a better matchup—both have 4-core ARM CPUs and a more limited PCI Express Gen 2 bus at their heart.

Around the same time, Radxa also sent me their new Taco—a less-than-$100 Raspberry Pi Compute Module 4 carrier board with 5x SATA ports, 1 Gbps and 2.5 Gbps Ethernet, an M.2 NVMe slot, and an M.2 A+E key slot. (The Taco will soon be available as part of a kit with a CM4 and case for around $200.)

The specs evenly matched, at least on paper:

But specs are one thing, measurable performance is another.

Disk performance

I benchmarked the raw disk access performance with fio and iozone to get a general idea of how fast the SATA drives would perform in RAID 5.

I chose RAID 5 because it taxes all the subsystems that are traditionally weak points on lower-powered ARM boards: the SoC’s CPU for parity calculations when writing, the PCIe bus for throughput, and the SATA controller.

Note: All the details of my test methodology and benchmarks I ran are documented in this GitHub issue.

Here’s how raw disk performance looks:

The Pi is faster for reads, but the ASUSTOR somehow wipes the floor on writes. Seeing that the Pi board also had space for an NVMe drive, I also set up bcache in Pi OS in writeback mode to give the Pi a boost. And that definitely helped:

But NASes have to expose raw storage to a network—and that’s an area lower-end NASes often fall short, especially when they try saturating more than a 1 Gbps network connection!

Network and Samba performance

Indeed, both the Taco and the Drivestor seemed to struggle to saturate a 2.5 Gbps network connection in their default configuration—both using the Realtek driver built into the Linux kernel:

But I noticed if I switched the Pi over to Realtek’s own driver, I could fully saturate the connection:

I wrote up a blog post about the driver issue, but it seems like Realtek’s driver has some optimizations that offload a lot of the network packet processing from the CPU, or somehow saves on interrupts by a very significant amount.

But that network throughput doesn’t automatically translate to network file copy performance, as demonstrated in my next benchmark:

The benchmark above was a large file copy—the best case scenario. And both the Pi and the Drivestor were very consistent across multiple tests. When you tax a low power SoC with RAID 5 and set it up as the traffic cop between SATA and a 2.5G Ethernet port, it’s obvious the performance is more limited than more expensive Intel/AMD options.

Overclocking could help, but honestly, if you want to see supercharged network file copy performance, opt for a more expensive and better endowed NAS.