File Download speed/max connections

[speedingcheetah 39]

I share files via Emby, meaning I have users that use the Download option. (they connect via HTTP link)

I have 1gig speed internet (both download and upload).

Is there some sort of speed cap, or connection limit with Emby server?

One user, now has 1gig fiber internet as well. (used to have 40megs DSL)

He only gets about 6MB's downloading files from my Emby server, using FireFox default  file downloader. 

(he was getting about 3MB's with his old internet).

I  first thought that FF built in downloader was just one connection and was slow, but i just tested it downloading a 3GB iso file of Kail Linux and it pulled about 100MB's...so its not FF that is limiting users download from Emby. 

I am in MN, he is in NH, so there is a fairly large distance between us, however, I don't see that being such a restriction.

Anything I can look at/tweak to make the downloads faster?

Server is currently 4.5.0.20 beta running on a Intel NUC i7  Linux Mint 19.3 and has a 10Gig NIC (via Thunderbolt)

[Q-Droid 634]

If you connect to your own public Emby URL to download, preferably from a different client machine, what speed do you get? Are you using reverse proxies, Cloudflare, firewalls, filters, etc?

 

[speedingcheetah 39]

12 minutes ago, Q-Droid said:

If you connect to your own public Emby URL to download, preferably from a different client machine, what speed do you get? Are you using reverse proxies, Cloudflare, firewalls, filters, etc?

 

On the same LAN, i get full wired speeds.

It has been a while since i tried on a remote connection, the only one available to me is the free Xifnity wifi hotspot, and that max speed is only about 30Mbps, and I just tested that and I get about 3MB's download. (which is max bandwidth for that connection)

Long time ago, i did test at my Grandparents home, when they had fast Comcast cable internet, 250+ and i was getting about 20MB's then, but that was like 2yrs ago or so, since i tried, they now have much slower speed due to cost.

I do not have any other very high speed remote internet to test with at this time.

No, firewals, prxoies or anything fancy.

The same sever computer can download and upload files to and from file shares site, like Gdrive, or MEGA, at 50MB's+ easy. 

Only Emby is being slow (as reported by users)

I put the same file on GDrive and same user can download it at his 1gig speeds. 

Edited August 22, 2020 by speedingcheetah

[Q-Droid 634]

What I'm asking is that you test your public URL connection from within your LAN (haripin/loopback). Is that what you did? The idea is to see if Emby is somehow throttling public connections, ie: treating those differently from local connections. If you did that already and don't see a difference then the slowness is likely between your CPE and your remote user, not Emby.

You could also run a file transfer test outside of Emby between you and your remote user. A very basic HTTP server can be setup with python from the command line:

Place a large file in a directory.
cd to that dir.
Shutdown Emby (temporary).
python -m SimpleHTTPServer 8920
Ask the remote person download the file using HTTP, not HTTPS.
Ctrl-C out of python.
Start Emby.

I'm assuming your Emby public IP is using the HTTPS port, if not change the python line to 8096.

 

[speedingcheetah 39]

delete post.

Edited August 22, 2020 by speedingcheetah

[speedingcheetah 39]

Accessing the Emby sever, via the HTTP local IP address, or via the Public URL, results in the same speed for downloading,  ~105MB's. it goes instantly to that speeds, and holds there.

I did have a FTP sever on the thing while back, and that did not have any such limits, also, i have done remote iperf3 tests with some one from Ubiquiti, as part of a different thing...

(New post, my first test, i forgot I was on wifi....)

Edited August 22, 2020 by speedingcheetah

[Q-Droid 634]

In that case I would say the issue is beyond Emby. A drop to 6MB is (~50mbps) is pretty big so a raw file transfer test might be worth a try. Also, if you can, run some remote tests over different ports though this means changing your port forwarding setup.

 

 

[speedingcheetah 39]

Just now, Q-Droid said:

In that case I would say the issue is beyond Emby. A drop to 6MB is (~50mbps) is pretty big so a raw file transfer test might be worth a try. Also, if you can, run some remote tests over different ports though this means changing your port forwarding setup.

 

 

Thanks for you opinion.....but I am looking for a response from the devs of Emby, for any thing I can do in the server software it self. or logs.

[speechles 1917]

You have a 0 in the upload speed limit and saved?

Also it may be converting at the same time as downloading depending on device. Notice if it is converting it will do it at full speed with the box checked. This means it cannot send it at your full bitrate if it is converting it can only send as fast as it converts.

Edited August 22, 2020 by speechles

[speedingcheetah 39]

2 minutes ago, speechles said:

 

You have a 0 in the upload speed limit and saved?

I have no value in there at all. never touched that setting.

 

i ran a tracert to users ip addres, to see if there was any sort of high ms or latency, look pretty good.

 

I have sent files to other countries with far higher numbers, and still have speeds of 80MB's.

Edited August 22, 2020 by speedingcheetah

[speechles 1917]

Are you bottle necked to that user IP->IP ?

What if you send a file to that person circumventing Emby. Just directly send them something via some other means. Is this happening at full speed or also bottle necked in the same way?

tracing the route and actually sending packets across the route are different. Trace will tell you which server among the hops is being problematic. But it won't spot a bottle neck unless it is shoveling massive amounts of data across the pipe.

 

https://www.solarwinds.com/network-bandwidth-analyzer-pack/use-cases/network-bandwidth-bottleneck

Above offers a free trial and can be used to see where faults are.

Edited August 22, 2020 by speechles

[speedingcheetah 39]

5 minutes ago, speechles said:

Are you bottle necked to that user IP->IP ?

What if you send a file to that person circumventing Emby. Just directly send them something via some other means. Is this happening at full speed or also bottle necked in the same way?

tracing the route and actually sending packets across the route are different. Trace will tell you which server among the hops is being problematic. But it won't spot a bottle neck unless it is shoveling massive amounts of data across the pipe.

I have seeded a temp private torrent to that user, and that went to about 90MB's after a few minutes. Though torrent is not same as HTTP..soo. not sure if that counts.

EDIT: also yes, i have used Filezilla and send files at at least 30MB's to them and others, via the same machine.

Edited August 22, 2020 by speedingcheetah

[speechles 1917]

Torrent is p2p and you have a large number of peers drawing from you. With downloading it is a single user to a single user. Not the same as torrents unless you had only a single peer. How many peers were on your torrent downloading from you? That can also cause the issue is one fellow with a very fast network with 0 bottle necks can crush other peers downloading from you. One super fast guy close to you can crush the speeds of several very far from you. Thereby crushing their speeds to near zero while that single close person enjoys incredible super fast speeds. But this is momentary. It is an overload technically when people get crushed speeds. It won't occur all the time. If you find this always occurs it is most definitely a bottle neck and not just overloading of nodes with peers.

https://morningconsult.com/opinions/competition-for-bandwidth-during-covid-19-is-crushing-america-5g-can-help/

This is likely the cause of the bottle neck. Some companies are purposely slowing speeds to prevent overloading nodes because so many are at home during COVID-19.

Edited August 22, 2020 by speechles

[speedingcheetah 39]

12 minutes ago, speechles said:

 

tracing the route and actually sending packets across the route are different. Trace will tell you which server among the hops is being problematic. But it won't spot a bottle neck unless it is shoveling massive amounts of data across the pipe.

With user previous connection, there was one hop that had 450ms, and 2 that were over 200ms....so this a big inprovement

[speedingcheetah 39]

he gets fasts speeds all the time, from litterly everything else. my emby server the only thing that is slow for him. not sure how much more clear i can be.

and this not a recent thing, he always has gotten slow speeds from me for years, even back when i had slow DSL.

we both have un filtered, un capped direct line fiber now. it is not a shared connection.

[Q-Droid 634]

2 minutes ago, speedingcheetah said:

With user previous connection, there was one hop that had 450ms, and 2 that were over 200ms....so this a big inprovement

You can't rely on traceroute for this. The only connection that matters is the final destination and the routers in between often ignore or deprioritize ICMP responses.

One reason I suggested raw file transfers using other ports beside the Emby ports was to see if there's any traffic shaping going on. I'm the last person you'd hear claim ISP throttling or traffic shaping but it's not beyond the realm of possibility.

 

[speedingcheetah 39]

26 minutes ago, Q-Droid said:

You can't rely on traceroute for this. The only connection that matters is the final destination and the routers in between often ignore or deprioritize ICMP responses.

One reason I suggested raw file transfers using other ports beside the Emby ports was to see if there's any traffic shaping going on. I'm the last person you'd hear claim ISP throttling or traffic shaping but it's not beyond the realm of possibility.

 

Is there a  website user can go to, to run a test to see if there is throttling or shapeing happeing on certain ports?

is there a good port to use other than 8096?  

how about 9090?  does it really matter?

edit: looked up on a random site, that shows ports assigned to things, 8092-8096 are unasigned. where as, say 9090 has many things listed for use.

is it best to use unasigned ports? or more used ports?

Edited August 22, 2020 by speedingcheetah

[Q-Droid 634]

I haven't tried to find sites or tools to test for shaping. Like I said it's not my goto reason for throughput issues.

You can pretty much use any port that's available on your system and higher than 1023. Assigned generally refers to well known service ports used for consistency and likely to be in use. You'll find these in /etc/services. It's a soft reservation of the default for lack of a better term. You can run netstat -lnp on your system to see what's in use.

This is about testing to identify and isolate the bottleneck, not for permanent use. Good range to try is between 7000-9999. The idea is to try a port that is not well known for the type of service likely to be throttled or shaped. For example, well known P2P ports are likely to be targets. On the other hand 80 and 443 are what most user traffic on the internet use. You could try forwarding 80 -> 8096 and/or 443 -> 8920 if your router allows it.

 

 

[Q-Droid 634]

2 hours ago, speedingcheetah said:

Accessing the Emby sever, via the HTTP local IP address, or via the Public URL, results in the same speed for downloading,  ~105MB's. it goes instantly to that speeds, and holds there.

I did have a FTP sever on the thing while back, and that did not have any such limits, also, i have done remote iperf3 tests with some one from Ubiquiti, as part of a different thing...

(New post, my first test, i forgot I was on wifi....)

Can you share your server log from this test? I want to make sure the server saw this connection as remote rather than local. Or you can verify this yourself if you wish.

 

[speedingcheetah 39]

I have tried changing ports, tried a few, 8089, 8999, 65525...any port other than the default 8096 result in a slight decrease to about 8MB's. Change back to default, its back up to 12-15MB;s.

EDIT: I also put file on Gdrive, and send link, and he gets 25MB's. way faster.

Edited September 12, 2020 by speedingcheetah

[speedingcheetah 39]

As a test, i installed JellyFin,  on my old  and ancient laptop, and set port forward to that system,  that resulted in a download of 18-24MB's. 

Went back to Emby on my main system, 8MB's.

so, something is up with Emby, the OS, the NIC,....that unit configuration and how it it is working with Emby and its connections..????

The OS on the Intel NUC that is my sever, is Mint 19.3, fully up to date. the one on the old laptop is Mint 20.

Using the integrated NIC on the laptop, vs the Thunderbolt 10Gig NIC i have connected to NUC.

Can't be hardware issue, as I can push my full 1gig upload through the thunderbolt NIC, via torrents, uplaods to Gdrive or other servers that are outside my LAN,

OS version?? maybe?  not about to wipe and re-do the thing.

something weird with Emby, idk.  

Not about to install JellyFin on same system, as they use the same ports and, well, J.F is Emby, just someones clone/fork of it. that i dont care much for.

I have Channels DVR server also running on the Intel NUC, but that is not setup for Remote Access, and is not using any cpu or bandwith durring these tests.

I may try using the onboard NIC on the NUC, see if that affects things, but,  not an option for me to loose the 10Gig connection to that system. so kinda  mute point to me.

 

Edited September 12, 2020 by speedingcheetah

[speedingcheetah 39]

So. i did try using the internal 1gig NIC, and that made the download speeds to remote user jump to , started out about about 45MB's but slowly dropped to about 20MB's....but fastest yet.

back to using the Thunderbolt 3 NIC, and its capped to 10MB's.

I don't get it. has to be something with Emby and this hardware?  as i have used the T3 NIC to upload files to G drive, other sites, torrents,  and to other peoples ftp servers, pushing 80+MB's for hours.

Only remote Emby users are stuck at 10MB's when using that NIC.

Something in the Linux Kernel?....idk.

at this point, no idea what else to do.

Edited September 13, 2020 by speedingcheetah

[Luke 37019]

Perhaps the limit is on their end.

[speedingcheetah 39]

14 minutes ago, Luke said:

Perhaps the limit is on their end.

Clearly, it is not. as proven by my testing.

[speedingcheetah 39]

15 minutes ago, Luke said:

Perhaps the limit is on their end.

As my recent testing to narrow things down seems to indicate, it is a very weird and specific scenario.

Emby, somehow, is the problem.

This is only the case with Emby and what the narrowed down variable has become, the NIC used on the machine.

------

Emby when used with the Thunderbolt 3 10Gig NIC i have, is limiting remote users to ~9MB's max.

When use the on-board Intel 1gig NIC, remote users get ~45MB's - 20MBs. (maxes out their ISP download speed, then it slows down)

-----

(My normal config, has been for at least 3yrs, to use the Thunderbolt 3 10Gig NIC...maybe this remote speed limit been a problem since i fist use it, it was not until recently that 2 of the main users that download  videos from me got faster than 100meg internet connections. They only had 50meg packages before.)

I can push ~950megs upload all day to any other site/service....Gdrive, torrent, ftp, online backup server...i have done so for several days constant transfer, not had one single issue or drop on speeds,  and local file transfers to and from the computer are fast as hell still, so there is no hardware fault of the NIC.

So the question for you devs...how/why would Emby perform so different, for remote connections only, among 2 different Network Interfaces hardware, when nothing else on the system exhibits the same behavior.