After publishing our article about Microsoft’s opt-in native NVMe function on Home windows Server 2025, we acquired a number of requests for a direct comparability of storage efficiency between Home windows Server 2025 with native NVMe and a Linux-based server OS. One particularly enthusiastic Redditor even supplied us beer to do it! Since there have been clearly no different cheap choices, we determined to run the identical exams on Linux.
A Lengthy Time In the past, in an OS Model Far, Far Away
The Linux kernel has supported NVMe since model 3.3, launched in March 2012. Equally, the protocol has been supported on Home windows Server (non-natively, by way of SCSI translation) since 2012 R2, round October 2013. Greater than a decade later, customers are nonetheless debating whether or not Home windows or Linux is healthier for storage, so we thought we’d add some extra gasoline to the hearth with benchmark outcomes evaluating the 2.
Since we now have take a look at outcomes for Home windows Server 2025 utilizing each non-native and native storage stacks, we thought it was applicable to judge two storage stacks on Linux. For our FIO benchmarks, we used each libaio and io_uring, two of the most well-liked APIs for storage transactions. Whereas io_uring is significantly newer and gives many enhancements for asynchronous I/O, libaio stays broadly used for its flexibility and ease of use (Didona, Pfefferle, Ioannou, Metzler, & Trivedi, 2022). Full structure overviews for each stacks are past the scope of this text, however we’re nonetheless offering outcomes for a direct comparability.
Testing NVMe on Ubuntu Server 24.04.4 LTS
Our {hardware} platform for this comparability is identical server utilized in our Home windows Server 2025 native NVMe article. To make sure most throughput and constant outcomes, it’s outfitted with two 128-core AMD EPYC 9754 CPUs, 768GB of DDR5 reminiscence at 4800 MT/s, and fifteen 30.72 TB Solidigm P5316 NVMe SSDs with PCIe 4.0 in a JBOD configuration.
As we talked about in our earlier article, the Solidigm P5316 has an indirection unit dimension of 64 kilobytes, which implies write efficiency for smaller sizes (resembling 4K exams) is commonly worse than anticipated. As soon as once more, we ran completely different take a look at patterns at block sizes of 4K, 64K, and 128K to offer a broad vary of outcomes for learn and write operations.
We chosen Ubuntu Server 24.04.4 LTS as our Linux instance due to its recognition and long-term help. It runs on the Linux kernel 6.8 by default, which isn’t the latest or most superior, however possible represents a big portion of installations worldwide.
Highlights
- Home windows Server 2025 native NVMe wins in three out of 4 learn efficiency benchmarks
- Decrease CPU utilization noticed with Home windows Server throughout most exams
- Ubuntu Server 24.04.4 LTS wins in three out of 4 write efficiency benchmarks
| Metric | Random 4K | Random 64K | ||||||
|---|---|---|---|---|---|---|---|---|
| Home windows Non-Native | Home windows Native | Linux libaio | Linux io_uring | Home windows Non-Native | Home windows Native | Linux libaio | Linux io_uring | |
| Random Learn | ||||||||
| Bandwidth (GiB/s) | 6.1 | 10.058 | 9.198 | 9.504 | 74.291 | 91.165 | 77.517 | 77.7 |
| IOPS | 1,598,959 | 2,636,516 | 2,411,000 | 2,491,000 | 1,217,176 | 1,493,637 | 1,270,000 | 1,273,000 |
| Common latency (ms) | 0.169 | 0.104 | 0.198 | 0.192 | 0.239 | 0.207 | 0.377 | 0.376 |
| Complete CPU Utilization (%) | 72.67 | 74.22 | 99.77 | 99.76 | 68.44 | 65.11 | 83.16 | 84.72 |
| Metric | Sequential 64K | Sequential 128K | ||||||
|---|---|---|---|---|---|---|---|---|
| Home windows Non-Native | Home windows Native | Linux libaio | Linux io_uring | Home windows Non-Native | Home windows Native | Linux libaio | Linux io_uring | |
| Sequential Learn | ||||||||
| Bandwidth (GiB/s) | 35.596 | 35.623 | 31.867 | 31.433 | 86.791 | 92.562 | 97.05 | 97 |
| IOPS | 583,192 | 583,638 | 522,000 | 515,000 | 710,978 | 758,252 | 795,000 | 795,000 |
| Common latency (ms) | 0.809 | 0.812 | 0.919 | 0.932 | 0.613 | 0.608 | 0.603 | 0.604 |
| Complete CPU Utilization (%) | 44.89 | 37.11 | 53.94 | 41.74 | 61.56 | 49.56 | 75.14 | 76.90 |
| Metric | Random 4K | Random 64K | ||||||
|---|---|---|---|---|---|---|---|---|
| Home windows Non-Native | Home windows Native | Linux libaio | Linux io_uring | Home windows Non-Native | Home windows Native | Linux libaio | Linux io_uring | |
| Random Write | ||||||||
| Bandwidth (GiB/s) | 1.803 | 1.756 | 1.876 | 1.815 | 7.654 | 7.655 | 7.652 | 7.651 |
| IOPS | 472,725 | 460,383 | 492,000 | 476,000 | 125,391 | 125,406 | 125,000 | 125,000 |
| Common latency (ms) | 0.992 | 1.028 | 0.974 | 1.007 | 3.814 | 3.816 | 3.827 | 3.828 |
| Complete CPU Utilization (%) | 26.00 | 20.67 | 45.76 | 22.80 | 12.22 | 9.33 | 20.07 | 10.90 |
| Metric | Sequential 64K | Sequential 128K | ||||||
|---|---|---|---|---|---|---|---|---|
| Home windows Non-Native | Home windows Native | Linux libaio | Linux io_uring | Home windows Non-Native | Home windows Native | Linux libaio | Linux io_uring | |
| Sequential Write | ||||||||
| Bandwidth (GiB/s) | 44.67 | 50.087 | 52.283 | 52.25 | 50.477 | 50.079 | 52 | 52.083 |
| IOPS | 731,859 | 820,603 | 856,000 | 856,000 | 413,495 | 410,232 | 426,000 | 427,000 |
| Common latency (ms) | 0.399 | 0.558 | 0.560 | 0.560 | 1.022 | 1.149 | 1.126 | 1.125 |
| Complete CPU Utilization (%) | 70.44 | 57.78 | 61.88 | 62.75 | 58.44 | 47.33 | 61.49 | 44.27 |
Observe: Our Linux IOPS outcomes are rounded to the closest thousand attributable to variations in FIO reporting between Home windows Server 2025 and Ubuntu Server 24.04.4 LTS. Bandwidth, latency, and CPU utilization outcomes are rounded constantly throughout each platforms.
The Numbers Don’t Lie
Instantly, we see that Ubuntu doesn’t outperform Home windows in each class. Whereas libaio and io_uring delivered wonderful throughput in our random-read bandwidth exams, they didn’t match the efficiency of Microsoft’s native NVMe stack. The Home windows NT kernel beat the Linux kernel by about 17% in our random-read 64K exams, with a profitable 91.165 GiB/s on native NVMe versus io_uring’s better of 77.7 GiB/s.
Nonetheless, not all hope is misplaced for Torvalds’ technological terror. Ubuntu Server narrowly beat Home windows Server in one in all our learn efficiency benchmarks: the sequential 128K take a look at. Right here, Linux’s libaio carried out greatest at 97.05 GiB/s, in comparison with Home windows’ native NVMe at 92.562 GiB/s, a distinction of about 5%. This means that Linux might maintain a slight edge when managing block sizes bigger than the drives’ indirection models.
Random write bandwidth was constant throughout each Linux and Home windows, particularly in 64K benchmarks. The perfect and worst outcomes from these exams differed by solely 0.05%, suggesting that each one storage stacks realized the drives’ full potential.
Apparently, the Linux 6.8 kernel claimed victory in sequential write bandwidth exams for block sizes of 64 Okay and 128 Okay. Whereas the distinction was not huge, the open-source software program stacks beat Home windows Server’s native NVMe by about 2 GiB/s in each instances.
Latency outcomes usually adopted throughput take a look at outcomes, proven greatest by the distinction in random learn averages. Sadly for Tux, libaio and io_uring had larger latency, with the biggest distinction of 0.17 ms between Home windows Server native NVMe (0.207 ms) and libaio (0.377 ms) for 64 Okay random reads.
Maybe probably the most stunning revelation from our benchmarks is the huge delta in CPU utilization between Home windows Server 2025 and Ubuntu Server 24.04.4 LTS. In three of 4 random and sequential-read benchmarks, Home windows Server native NVMe had the bottom CPU utilization. Essentially the most notable outcome was seen in the course of the sequential learn 128K benchmark run, the place Home windows used 27.34% lower than Linux.
CPU utilization with libaio and io_uring carried out barely higher in random and sequential write exams, but it surely was nonetheless not sufficient to stop native NVMe on Home windows Server from profitable in three of the benchmarks. A notable exception was libaio’s CPU utilization in the course of the random write 4K take a look at, which reached 45.76% of the system’s CPU, whereas the opposite storage stacks hovered round 20%.
Winner Winner, CPU Dinner
As our outcomes present, Home windows Server and Ubuntu Server carry out carefully when examined head-to-head in each random and sequential efficiency exams at completely different block sizes. When it comes to bandwidth, Home windows Server 2025 with native NVMe usually outperformed Linux in most learn exams, whereas Linux responded with barely higher ends in our write exams. Our latency figures informed the same story, however the true spotlight was Home windows Server 2025’s CPU effectivity when utilizing native NVMe.
Microsoft has clearly put effort into making their latest storage stack their greatest, and whereas it doesn’t at all times win towards libaio and io_uring, it places up combat. Whereas these outcomes are usually not definitive throughout all use instances and server setups, they might assist server directors resolve whether or not to deploy a Home windows or Linux server when storage efficiency is extra essential than OS compatibility.
Tell us what you consider these outcomes by commenting on our social platforms or the SR Discord! Did you count on Home windows Server to take action properly in our exams, or had been you rooting for Linux? Would you wish to see extra Linux server distributions or kernels examined? We’re at all times in search of your suggestions, and reader-requested exams like these typically turn into our favourite articles.
References
Didona, D., Pfefferle, J., Ioannou, N., Metzler, B., & Trivedi, A. (2022, June 13). Understanding Trendy Storage APIs: A scientific research of libaio, SPDK, and io_uring. SYSTOR ’22, 120-121. Retrieved April 3, 2026, from https://atlarge-research.com/pdfs/2022-systor-apis.pdf
