Is ~55MB/s "reasonable" over AC?

[turnstyle]

Hi all, I'm in the process of getting up to speed some new network gear.

My question regards a test file transfer from Windows PC, wired by Ethernet to a new Archer C8 (AC1750), and then over AC to a Mac.

For the test I created a share on the PC (fwiw, it's an SSD) and put two 1GB files in it.

On my Mac (and over the AC wifi, sitting near the router) I connected to that share via a smb: path, and then in Finder I copied those two files from the shared folder on the PC to my Mac.

This copy (two files totaling 2GB) took about 35 seconds -- so roughly 55MB/s.

Is that about what I should expect?

I thought AC over 5GHz was spec'ed at 1300Mbit/s (or ~162MB/s) -- I don't expect to reach that theoretical limit, but is 55MB/s "to be expected"?

Thanks kindly!

 

[thiggins]

Didn't we already put this one to bed in your other thread?
http://www.snbforums.com/threads/just-upgraded-from-n-to-ac-reality-check-mac-to-pc-backup.26859/

 

[turnstyle]

Didn't we already put this one to bed in your other thread?
http://www.snbforums.com/threads/just-upgraded-from-n-to-ac-reality-check-mac-to-pc-backup.26859/

Nope!

That was about performance of a connection over AC to a PC with an external USB2 drive -- and from what I gathered, USB2 was the bottleneck.

I then realized I could set up a test share on the *internal* drive of the PC (an SSD) to eliminate the USB2 bottleneck.

So, my understanding is that 35MB/s is "reasonable" for copying a big file from a USB2 drive over AC.

And now I'm trying to find out if 55MB/s is "reasonable" having eliminated the USB2 bottleneck.

I was kind of expecting it to be faster, given then 1300Mbit AC spec -- "theoretically" about 162MBs.

For example, if I get a Synology or QNAP, am I still going to get roughly 55MB/s over AC?

Many Thanks... -Scott

 

[thiggins]

You will never achieve a maximum link rate due to protocol overhead.

This chart shows maximum throughput we have obtained with our 3x3 AC1900 class test client in 5 GHz is around 600 Mbps, with a VERY efficient test stream.

You should also take a Gigabit Ethernet connected baseline measurement for a reference.

 

[Caple]

Well, the chart for the Archer C8 states a 481 Mbps throughput. That's ~60MB/s, right?
If the values were obtained on a VERY efficient test stream I believe 55MB/s is ok.
Am I wrong?

 

[turnstyle]

Thanks so much, I wasn't aware of that chart!

So now I gather the ~440Mbit/sec I'm getting is reasonably in line with expected real-world AC performance, per your chart. I guess I'm surprised to learn despite the 1300Mbit/sec description of AC, that only 600Mbit/s is actually possible (and even then, only with a "VERY" fine-tuned set-up) -- regardless, it's good to know I'm roughly where I should be. And I do understand marketers will describe products to give favorable impressions, even if I personally would prefer they describe product with more useful metrics. All's good.

So then one last question:

I had been thinking upgrading my file server to use USB3 drive -- or switching to a NAS -- would roughly triple my speed -- from my current USB2 bottleneck of 35MB/sec.

But given the "real world" bottleneck of AC1750 is roughly 55MB/sec -- if I get a NAS or switch from USB2 to USB3 -- I'm still only improving from 35MB/sec to 55MB/sec -- is that about right?

If so, then I might actually skip the NAS for now, and simple stick to my USB2 drives (which is a pretty surprising outcome!). I do understand that if all my clients were wired, then USB3/NAS would be a big step up. But my clients are all wireless.

Right? If I'm "mostly on wifi" then I'll only ever get up to 55MB/sec over AC -- and if I'm currently seeing 35MB/sec over AC to USB2, the best I'll ever see to a NAS or USB3 is still constrained to 55MB/sec?

This is tremendously helpful! (And I think you're saving me from wasting some money!)

-Scott

 

[L&LD]

Caple, you're right.

Transfer rates in MB/s vary depending on the number, size and type of the file(s) transferred.

They also vary depending on the protocol used (USB, Ethernet or other).

Another factor is the source device too (hdd, usb stick, ssd).

If transferring from wireless to wireless (unless on different bands), then that effectively halves the throughput too.

The chart Tim showed indicates about 74MB/s download for the EA8500 at it's max. Average is about 39MB/s and for that particular router, just over 2MB/s as it's minimum (even the best tested router was just over 10MB/s using a different testing procedure).

What that means is that if you are achieving close to those speeds, depending on the test and devices you use, you should be satisfied with being in the upper half of that range.

That 162.5MB/s theoretical limit depends on three antennae, three stream clients in a zero interference environment that are setup very close to the router tested and do not take into account the encoding losses for different connection types.

As you can see, that type of setup is so theoretical when all is taken into consideration that reality at it's best with the EA8500 is only achieving about 45% of what the formulae would indicate.

 

[L&LD]

Thanks so much, I wasn't aware of that chart!

So now I gather the ~440Mbit/sec I'm getting is reasonably in line with expected real-world AC performance, per your chart. I guess I'm surprised to learn despite the 1300Mbit/sec description of AC, that only 600Mbit/s is actually possible (and even then, only with a "VERY" fine-tuned set-up) -- regardless, it's good to know I'm roughly where I should be. And I do understand marketers will describe products to give favorable impressions, even if I personally would prefer they describe product with more useful metrics. All's good.

So then one last question:

I had been thinking upgrading my file server to use USB3 drive -- or switching to a NAS -- would roughly triple my speed -- from my current USB2 bottleneck of 35MB/sec.

But given the "real world" bottleneck of AC1750 is roughly 55MB/sec -- if I get a NAS or switch from USB2 to USB3 -- I'm still only improving from 35MB/sec to 55MB/sec -- is that about right?

If so, then I might actually skip the NAS for now, and simple stick to my USB2 drives (which is a pretty surprising outcome!). I do understand that if all my clients were wired, then USB3/NAS would be a big step up. But my clients are all wireless.

Right? If I'm "mostly on wifi" then I'll only ever get up to 55MB/sec over AC -- and if I'm currently seeing 35MB/sec over AC to USB2, the best I'll ever see to a NAS or USB3 is still constrained to 55MB/sec?

This is tremendously helpful! (And I think you're saving me from wasting some money!)

-Scott

I think you're finally beginning to understand.

But, your conclusions are wrong. USB 2.0 or 3.0 is still a poor substitution for a NAS with a 1GBe LAN connection (even if all your clients are wireless).

The issue is how the USB protocol comes to a crawl with small files. If you're only transferring huge files wirelessly, then you may be able to get away with using USB enclosures off the router.

If your file transfers are more real world and varied, then the NAS will offer performance greater than the about 60% increase the numbers would indicate.

I implemented a simple backup process for a customer recently and the initial transfer over WiFi took over 7 hours but that was expected as the data set was large (about 30GB). What was unacceptable was when I returned and about 250MB of changed data was backed up again. That took over an hour to complete.

When I asked them again if I could see the NAS they were using, they finally admitted it was a USB drive they were backing up to. The real issue? Those 250MB's of changed data was over 8,200 files, some of them very, very tiny (less than 68 bytes).

With the inefficient USB protocol that tiny amount of data brought their workflow to a standstill (at least when doing backups). Upgrading to a TS-131 single bay NAS and using their USB drive to backup the NAS brought that daily backup routine to less than 3 minutes.

And; they had two backup's of their data too.


For your workflow, you may be satisfied with your current level of performance. But even with only large file transfers, you will be saving about 57% time doing wireless transfers. That adds up very fast to making the $200 price of a NAS like the TS-131 and the price of a 3.5" hdd seem insignificant very quickly. Especially after you've experienced it for a short while.

While I do not normally recommend a single bay NAS, in your case (like in my customers example) it may be the easiest way to get multiple backups and higher performance too for the least cost.

 

[tim]

Best wireless speed I have ever seen is 550Mb/s. Asus RT-AC3200 to an Intel 7260 adaptor (about 5M apart).

 

[System Error Message]

You will never achieve a maximum link rate due to protocol overhead.

This chart shows maximum throughput we have obtained with our 3x3 AC1900 class test client in 5 GHz is around 600 Mbps, with a VERY efficient test stream.

You should also take a Gigabit Ethernet connected baseline measurement for a reference.

I managed to get faster by stressing everything to the point where it dropped some packets but it didnt involve the usual test methods you use. I used 2 computers with a very fast CPU and some good NICs and an ASUS AC68U. One was on wire while the other one wifi within a meter away. Average of 80% link utilisation using 2 stream AC wifi.

 

[turnstyle]

I think you're finally beginning to understand.

But, your conclusions are wrong. USB 2.0 or 3.0 is still a poor substitution for a NAS with a 1GBe LAN connection (even if all your clients are wireless).

The issue is how the USB protocol comes to a crawl with small files. If you're only transferring huge files wirelessly, then you may be able to get away with using USB enclosures off the router.

If your file transfers are more real world and varied, then the NAS will offer performance greater than the about 60% increase the numbers would indicate.

I implemented a simple backup process for a customer recently and the initial transfer over WiFi took over 7 hours but that was expected as the data set was large (about 30GB). What was unacceptable was when I returned and about 250MB of changed data was backed up again. That took over an hour to complete.

When I asked them again if I could see the NAS they were using, they finally admitted it was a USB drive they were backing up to. The real issue? Those 250MB's of changed data was over 8,200 files, some of them very, very tiny (less than 68 bytes).

With the inefficient USB protocol that tiny amount of data brought their workflow to a standstill (at least when doing backups). Upgrading to a TS-131 single bay NAS and using their USB drive to backup the NAS brought that daily backup routine to less than 3 minutes.

And; they had two backup's of their data too.


For your workflow, you may be satisfied with your current level of performance. But even with only large file transfers, you will be saving at least 11 minutes per GB transferred wirelessly. That adds up very fast to making the $200 price of a NAS like the TS-131 and the price of a 3.5" hdd seem insignificant very quickly. Especially after you've experienced it for a short while.

While I do not normally recommend a single bay NAS, in your case (like in my customers example) it may be the easiest way to get multiple backups and higher performance too for the least cost.

Thanks for that thorough and thoughtful reply!

One first question, you wrote -- "even with only large file transfers, you will be saving at least 11 minutes per GB transferred wirelessly" -- I don't understand this because right now I seem to be writing at 19MB/s to the USB2 over (using a single large file), so that's about 1 minute to write one GB. Just curious where 11 minutes per GB comes in?

So the notion I now need to work out is even though I'm seeing 35MB/s read and 19MB/s write over AC to the USB2 drive -- that ONLY applies to one big file -- and I need to try the same with 1GB of smaller files, that about right?

Does this "USB penalty" only apply when files are actually copied? Or does it also apply when my backup application is simply comparing local files to the files on the remote drive? (Even if it then decides the server copy is up-to-date?)

Lastly, the vendor of my backup software seems to think NAS is a potential problem for Mac users, do you happen to have any opinions on this?
www.econtechnologies.com/chronosync/TN-CS-why-your-nas-is-slow.html

Thank to everybody for chiming in!

 

[L&LD]

Sorry, that should be 57% not minutes (I'll correct my original post too that is 55MB/s to 35MB/s)...

The 'usb penalty' applies any time the disk is accessed for any type of action (read, write or compare).

The QNAP NAS's offer native mac support. I don't know how well it works.

But I can say this is similar to the small file issues I've discussed with you above (the metadata is a bunch of tiny files for every file you see/save).

Try with a 1GB folder of small files and see what kind of performance penalty there is. Test this wired and wireless and you'll know how much wireless encoding adds to the penalty too. Be sure you reboot between tests, so that the test folder is not cached and use a mix of files, not all the same type either (ideally; use a copy of your specific set of files).

Btw, the author of that article states that the topic will be approached in an objective way, but then goes on to placate the mac owners of how superior their hardware is.

If the NAS you pick (QNAP or Synology should only be considered, imo) doesn't do what you want; return it. Don't expect to fully understand if a possible solution will work by simply reading about it.

 

[turnstyle]

Thanks, here are my results transferring a 375MB folder of 7000 files:

Read from internal SSD over AC to Mac -- 53 seconds (tried twice)
Read from external USB over AC to Mac -- 2:07 the first time, 1:43 the second (and third) time
Write to internal SSD over AC from Mac -- 5:49
Write to external USB over AC from Mac -- 9:22

From this I conclude:

1) it takes a LOT longer to transfer a bunch of small files, regardless of whether I'm going through USB or to the internal drive.

* internal SSD drive read speed was 52MB/s for a big file -- but just 7MB/s for lots of little files

* external USB2 drive read speed was 35MB/s for a big file -- but just 3MB/s (first try) and 3.6MB/s (next try) for lots of little files

* internal SSD drive write speed was 31MB/s for a big file -- but just 1.1MB/s for lots of little files

* external USB2 drive write speed was 19MB/s for a big file -- but just 0.67MB/s for lots of little files

2) I do indeed see more benefit from the non-USB drive when it comes to lots of small files -- with big files, it was roughly 60% faster -- and with smaller files, it was roughly twice as fast.

So am I correct to expect a NAS would be roughly 60% faster than my USB2 drive for big files, and roughy twice as fast for small files? (I suppose twice is still good, but not quite the blowout I had been expecting.)

Does this data make sense? Thanks a ton for helping me make sense of this! -Scott

 

[stevech]

your list, #1 is a long well know and common sense situation. Fixed overhead per file, big file, overhead becomes less important. Like shipping cost of a $2 item.

proper throughput tests use software that move data between network devices/PCs using either TCP or UDP with test software on each end. No disk I/O. No SMB overhead. And max speed comes when there are 2+ streams (TCP) going simultaneously due to TCP's ACKs.

I would assume that SNB tests in this manner using Chariot or some such, not simple file transfers that color.

But yes, users want to know how long it takes now, with the new goodie, to transfer a 500MB file. How long to transfer 1000 small files is not improved much with a newer / faster goodie like 11AC devices.

 

[turnstyle]

your list, #1 is a long well know and common sense situation. Fixed overhead per file, big file, overhead becomes less important. Like shipping cost of a $2 item.

proper throughput tests use software that move data between network devices/PCs using either TCP or UDP with test software on each end. No disk I/O. No SMB overhead. And max speed comes when there are 2+ streams (TCP) going simultaneously due to TCP's ACKs.

I would assume that SNB tests in this manner using Chariot or some such, not simple file transfers that color.

But yes, users want to know how long it takes now, with the new goodie, to transfer a 500MB file. How long to transfer 1000 small files is not improved much with a newer / faster goodie like 11AC devices.

So overall -- if I'm currently using AC to communicate with a USB2 drive attached to a PC which is wired by Ethernet to the router -- switching to USB3 or getting a NAS will only improve my speed by about 60% for big files, and 100% for small files? Is that a reasonable expectation?

I was originally thinking I would see something like a 3x boost -- but if it's actually going to be more like 60% for big files and 100% for small files, that's good to know.

Anybody have anything to share about poor NAS performance with Macs? That was my other concern, for example, any of these links:
https://www.google.com/webhp?sourceid=chrome-instant&ion=1&espv=2&ie=UTF-8#q=mac+nas+slow

This has been tremendously informative!

 

[sfx2000]

So overall -- if I'm currently using AC to communicate with a USB2 drive attached to a PC which is wired by Ethernet to the router -- switching to USB3 or getting a NAS will only improve my speed by about 60% for big files, and 100% for small files? Is that a reasonable expectation?

If your worried about USB drive performance - hook the drives up to your Mac and be done with it - seriously... a USB hub is cheap enough...

Thing is, USB will not have the peformance of Firewire (as opposed to USB2) or Thunderbolt (as opposed to USB3), but for most folks, USB is good enough, and Mac's do USB quite nicely..

NAS peformance with Macs - I have not found that to be an issue - multiple Macs on my WLAN/LAN and I run a QNAP TS-453Pro, and no performance issues noted compared to Windows... SMB is a bit troublesome, but as long as AFP is also enabled, mounting the shares as AFP, I have performance that is "good enough" compared to a Native OSX server..

 

[turnstyle]

If your worried about USB drive performance - hook the drives up to your Mac and be done with it - seriously... a USB hub is cheap enough...

Thing is, USB will not have the peformance of Firewire (as opposed to USB2) or Thunderbolt (as opposed to USB3), but for most folks, USB is good enough, and Mac's do USB quite nicely..

NAS peformance with Macs - I have not found that to be an issue - multiple Macs on my WLAN/LAN and I run a QNAP TS-453Pro, and no performance issues noted compared to Windows... SMB is a bit troublesome, but as long as AFP is also enabled, mounting the shares as AFP, I have performance that is "good enough" compared to a Native OSX server..

The clients are all wireless (not desktops), so connectivity will be over AC.

Regarding your QNAP, and assuming you communicate with it over AC from one of your Macs, is this roughly what you get when copying via Finder:

* Reading a big fie from the QNAP to the Mac -- 52MB/s
* Reading a bunch of small files from the QNAP to the Mac -- 7MB/s
* Writing a big file from the Mac to the QNAP -- 31MB/s
* Writing a bunch of small files from the Mac to the QNAP -- 1.1MB/s

It would be VERY helpful to know if that's roughly what you experience when using your QNAP over AC from a Mac, when dragging files via Finer.

Thanks for taking time to share what you know!

 

[sfx2000]

I get around 700 Mbit over GigE across my NAS, about half of that on 802.11ac with a Macbook Air over WiFi (Macbook Air is a 2-stream 802.11ac device).

 

[sfx2000]

Big Files/Small files - it's the performance of the server, not the client - your performance issues are probably related to a desktop OS (which isn't meant to be a server), a slow Physical Layer Interface (USB2) and slow drives, which most USB drives are...

When going into a NAS box - the OS is optimized for filesharing, and in my case, I've got 4 SATA3 7200RPM drives running as a RAID10, which is fantastic for Read/Write performance...

 

[turnstyle]

I get around 700 Mbit over GigE across my NAS, about half of that on 802.11ac with a Macbook Air over WiFi (Macbook Air is a 2-stream 802.11ac device).

To be sure I follow, you're saying you get about 350Mbit over AC? So that seems roughly in line with 416Mbit (read) and 248Mbit (write) -- that is, if I am indeed understanding you correctly there.

It is possible to put a rough Mbit on your "large folder of small files" performance over AC? I'm particularly curious if you see something other than roughly 56Mbit (read) and 9Mbit (write)...