Multi-Client CIFS Performance for Consumer Workloads

The workloads experienced by a NAS unit in a typical home consumer setting have changed quite a bit over the last few years. Multiple mobile devices in a typical household raise the possibility that a NAS could be subject to the streaming out of multiple video files simultaneously. The popularity of IP cameras also make it necessary for NAS units to be able to record multiple video streams at the same time.

In our previous NAS reviews, we evaluated multi-client scenarios using synthetic workload traces and IOMeter. While there is nothing wrong in presenting numbers from such benchmarks, the reader is often left confused as to what those numbers might mean for his particular use-cases. Intel's NASPT benchmarking program gives us a good idea of the performance of the NAS unit when accessed by a single client. We took the source code of Intel's NASPT along with the supplied application traces and tweaked them to be able to run from more than one Windows client simultaneously in a co-ordinated manner. The graphs below present the results from tracking various metrics during the course of the benchmark runs. It must be noted that the average service times refer to what is obtained for all the traces when some of the data has already been cached in the client's memory. Unfortunately, NASPT doesn't provide any sort of guideline on what the optimal bandwidth and service times are for a good user experience.

Content Creation

The Content Creation workload seems to get acceptable performance for up to 3 clients. Beyond that, we have a drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

Content Creation - Multi-Client Benchmark

Folder Copy from NAS

The Folder Copy from NAS workload seems to get acceptable performance for up to 5 clients. Beyond that, we have a noticeable drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

Folder Copy from NAS - Multi-Client Benchmark

Folder Copy to NAS

The Folder Copy to NAS workload seems to get acceptable performance for up to 4 clients. Beyond that, we have a noticeable drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

Folder Copy to NAS - Multi-Client Benchmark

File Copy from NAS

The File Copy from NAS workload seems to get acceptable performance for up to 4 clients. Beyond that, we have a noticeable drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

File Copy from NAS - Multi-Client Benchmark

File Copy to NAS

The File Copy to NAS workload seems to get acceptable performance for up to 3 clients. Beyond that, we have a noticeable drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

File Copy to NAS - Multi-Client Benchmark

HD Video (1x) Playback

The HD Video (1x) Playback workload seems to get acceptable performance for up to 3 clients. Beyond that, we have a noticeable drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

HD Video (1x) Playback - Multi-Client Benchmark

HD Video(1x) Playback and Record

The HD Video(1x) Playback and Record workload seems to get acceptable performance for up to 3 clients. Beyond that, we have a noticeable drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

HD Video(1x) Playback and Record - Multi-Client Benchmark

HD Video (1x) Record

The HD Video (1x) Record workload seems to get acceptable performance for up to 4 clients. Beyond that, we have a noticeable drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

HD Video (1x) Record - Multi-Client Benchmark

HD Video (2x) Playback

The HD Video (2x) Playback workload seems to get acceptable performance for up to 3 clients. Beyond that, we have a noticeable drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

HD Video (2x) Playback - Multi-Client Benchmark

HD Video (4x) Playback

The HD Video (4x) Playback workload seems to get acceptable performance for up to 3 clients. Beyond that, we have a noticeable drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

HD Video (4x) Playback - Multi-Client Benchmark

Office Productivity

The Office Productivity workload seems to get acceptable performance for up to 5 clients. Beyond that, we have a noticeable drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

Office Productivity - Multi-Client Benchmark

Photo Album

The Photo Album workload seems to get acceptable performance for up to 3 clients. Beyond that, we have a drop in the per-client bandwidth numbers. The detailed table with a breakdown of all the throughput numbers as well as the service times is available here

Photo Album - Multi-Client Benchmark

The detailed logs from the processing of our benchmarks - inclusive of metrics such as the file open times for each workload on each of the clients - can be found here

Encryption Support Evaluation Multi-Client CIFS Performance for Professional Workloads
Comments Locked

22 Comments

View All Comments

  • Duncan Macdonald - Friday, September 25, 2015 - link

    Any NAS system that is limited to GbE or lower speed will give poor performance compared to even budget SSDs. (A GbE link can transfer about 100MB/sec after allowing for overheads - even low performance SSDs can do much better.) To beat locally mounted SSDs requires 10GbE or faster links. NAS systems are only useful for sharing files (slowly) to multiple computers or providing a backup far enough away to be unlikely to be affected by a common disaster (eg a house fire).
    As for NAS systems with 100Mb/sec links - AVOID (A USB 2.0 stick can be faster!!!)
  • BillyONeal - Friday, September 25, 2015 - link

    But most of the NASes here are well below saturating GigE. A USB 2.0 stick can be faster in extremely limited scenarios but in most cases USB protocol overhead per transfer will make it worse for these kinds of workloads.
  • Metaluna - Friday, September 25, 2015 - link

    Where in the article did anyone suggest using a NAS as a performance alternative to locally attached SSDs? And as for NAS only being useful for sharing files to multiple computers, yeah, that's kind of the whole point for why local area networks and file servers were developed in the first place. That's like saying "A GPU is really only useful for displaying images on your screen"
  • colinstu - Friday, September 25, 2015 - link

    don't know what 'overheads' you're talking about but my Synology NAS and gb network regularly transfer at 115MB/s (114-116). Still not the max theoretical of 125MB/s, but closer to the max then '100'
  • azazel1024 - Saturday, September 26, 2015 - link

    No, max theoretical is not 125MB/sec. That is raw data rate, but you can't actually transfer 125MB/sec of usable data over a 1GbE link. SMB max rate is about 117.5MB/sec using 9k jumbo frames and about 115MB/sec using standard 1500MTU. That is covering TCP/IP overhead as well as SMB overhead. Smaller file will reduce the max by a bit no matter how fast the host and server are because of additional SMB overhead involved in "opening" and "closing" each file transfer.

    NAS are just fine, at least newer moderately fast ones. But, I do have to say, if running windows based clients...a windows based server, if you can't/don't want to move to 10GbE can be significantly higher performing than a NAS, even in "undemanding" file transfers. My G1610 based server manages 235MB/sec between it and my desktop, both running Windows 8.1. Dual GbE NICs combined with SMB Multichannel is a beautiful thing.
  • UtilityMax - Sunday, September 27, 2015 - link

    NAS storage is slower than a directly attached storage! Shocking stuff! News at 11.

    GiE is is actually pretty acceptable for most applications, except a few specialist tasks. 10GbE can still be pretty expensive and power hungry.
  • UtilityMax - Sunday, September 27, 2015 - link

    Sorry mean 10GbE instead of GiE
  • Wixman666 - Sunday, September 27, 2015 - link

    So you decided that comparing apples and walruses is ok? A SSD and a 2 bay NAS have nothing in common for function, capacity, or price. Troll on, dude.
  • johnny_boy - Thursday, October 1, 2015 - link

    Any SSD system that is limited to SATA or even PCIE will give poor performance compared to even budget RAM disks. (A SATA link can transfer about blah MB/sec after allowing for overheads - even low performance RAM disks can do much better.) To beat locally mounted RAM disks requires bleek GbE or faster links. SSDs are only useful for reading and writing data.
    As for SSDs with blomps Mb/sec links - AVOID (A USB 3.0 stick can be faster!!!)
  • Wardrop - Friday, September 25, 2015 - link

    Do the btrfs snapshots show up in Windows under the "Previous versions" tab?

Log in

Don't have an account? Sign up now