Euphony OS w/Stylus player setup and issues thread

[austinpop]

For folks who are using Euphony OS, you may be interested in this. Željko now has an experimental build where he has implemented ramroot support, which can be enabled/disabled from the Expert Settings area of the UI.

 

I've been trying it out for a couple of days, and my results so far have been promising. Just like in AL, enabling ramroot (loading the root partition into a RAM disk, and then booting from it) results in a definite uptick in sound quality.

 

Željko wants to get a few more proof points before releasing it, and asked me to convey this call for testers. If you're interested in testing it out, and reporting your results here in this thread, please reach out to him by opening a support ticket in the Euphony system.

[austinpop]

58 minutes ago, mozes said:

So how can I  benefit from the ramroot option if I already run Euphony on an internal Optane drive?

Sorry if my question is a  bit dumb!

 

Not a dumb question, and no apology necessary! 

 

Loading the OS from an NVMe SSD (which is what your internal Optane drive is) is different from Ramroot:

• In normal mode, the boot loader will load the operating system (Euphony or AL, since both are based on Arch Linux) from the partition on the SSD that contains the root filesystem, i.e. that contains the / filesystem.
• In Ramroot mode, the boot loader proceeds with an extra step. Early in the boot process, it creates a RAM disk partition (/dev/zram0) in memory, copies the contents of the / partition from the SSD to this partition, and then completes the boot from this RAM disk partition. 

What this means is that the filesystem containing the kernel and the OS files is now completely in RAM.

 

Why exactly this sounds better is yet to be explained, although a lot of theories abound, having to do with reducing the latency of OS operations due to all OS files being in RAM.

[austinpop]

13 minutes ago, Chopin75 said:

But Euphony OS cannot run in RAMROOT correct ? I know AL can. 

 

Please read post #439.

[austinpop]

9 hours ago, misterspense said:

 

It was also once stated that the prime reason for using Optane was that the latency was comparable with RAM (and the noise comparable with USB). 

 

First, with regard to your question above. Optane's 3-D X-Point design may be faster than traditional SSD NAND, but it is still nowhere as fast as DRAM. For reference, see this table from Intel:

 

Source: https://www.theregister.co.uk/2016/04/21/storage_approaches_memory_speed_with_xpoint_and_storageclass_memory/

 

For this workload shown (not specified) Optane access is 35x slower than DRAM. Note that SSD is much slower - "...150 times slower than DRAM for writes and 600 times slower for reads." Obviously the exact numbers will vary with workload and different brands, but the basic point is clear - Optane's latency is still much higher than DRAM, while much lower than NAND SSD.

 

But all that said, this experiment being discussed here isn't related to the speed of Optane. It's to try ramroot mode on Euphony, just like it's been used on AudioLinux for all this while. Plenty of users are using ramroot on AL, while using Optane as their persistent storage. They still hear improved SQ.

 

The question being posed here, and what people testing the prototype are trying to determine with their ears is - is the same true with Euphony OS?

 

My ears tell me - yes.

[austinpop]

For those testing with Stylus, one other experiment to try is to see if caching - via the "Use Cache" flag makes a sonic difference when in ramroot mode.

 

I was unclear about how caching works, so got some more insight from Željko. Here are the salient points:

• the cache is maintained in the /data/Music/E_CACHE directory
  • NOTE: this directory is NOT loaded into RAM in ramroot mode, because it is on a separate disk partition
• on first play of a file, the stream is loaded from the original location, while in the background, the file is copied into the cache
• subsequent plays load from the copy in the cache
• The cache is allowed to grow until the free space in the /data partition falls below a threshold of some N GB. 
• After that any new additions to cache result in evictions using a standard LRU policy.

 

[austinpop]

Here is another update on Euphony OS with ramroot. I am running a standalone server.

 

The more I listen to ramroot with additional musical selections and genres, the more I am noticing how much it improves over the non-ramroot case on my system. The 2 scenarios I mostly run in my system are:

• Roon+StylusEP
• Stylus (no HQPlayer, no upsampling)

In general, while the benefit is quite significant on Roon+StylusEP, I feel the scenario that shows the biggest benefit is Stylus. This has caused the SQ gap between Stylus and Roon+StylusEP to increase further, which is honestly bittersweet for me, since I much prefer the Roon UI. Before the ramroot fix came along, I had gotten into a pattern of doing most of my listening on Roon+StylusEP, even though I knew Stylus sounded somewhat better.

 

Now the difference is just too big to ignore. I still listen to Roon+StylusEP for exploration, discovery, and if I'm "grazing," but for any sustained listening, I take the trouble to switch to Stylus and listen there.

 

I also tried to compare SQ in ramroot with "Use cache" enabled vs. disabled. This only applies to Stylus, as Roon Core does not interact with the cache. Also, when comparing, it's important to check that you are indeed listening to the cached version. The easy way to do that is to click on the album thumbnail on the bottom left of the screen, which opens up a full screen detailed view. One of the fields shows the location of the file. If a file is in cache, it will be in /data/Music/E_CACHE/...

 

At least to my ears, the "Use cache" case sounds better with ramroot, just like it did without ramroot.

 

Another caveat to note: my music files live on a NAS, so loading the original music file occurs over the network. It's possible that if your music lives on a locally attached disk, your findings may be different.

[austinpop]

18 minutes ago, mozes said:

Do you know if LMS+StylusEP interact with cache?

 

I doubt it, but don't know for sure.

[austinpop]

2 hours ago, HeeBroG said:

Hi All,

 

I finally managed to get Euphony running on my NUC7PJYH after using AL for some months.

Unfortunately I can only run it in Squeezelite mode as my Uptone LPS 1.2 is not powerful enough to power Stylus with a USB drive attached.

Still sounds good with Macmini running AL LMS as server.

Have 8GB of RAM in the NUC and have setup ram buffers as suggested previously on pg. 9 my Rajiv.

 

With LMS on the Mini, you should be able to run StylusEP on the NUC instead of SqueezeLite, and I find StylusEP to sound better. It also presents as a Squeezebox, just like SqueezeLite.

 

I also seem to remember that Željko implemented a distributed mode in Stylus. If so, you can also try running Stylus on Euphony on you Mini, and then StylusEP on your NUC. This would be better sounding than LMS, in my experience.

 

See what works and what you think.

 

2 hours ago, HeeBroG said:

Can someone help me setup CPU isolation for the NUC which has 4 cores.

 

Are you running with hyperthreading enabled or disabled? Either way, without entering any values, first just hit Apply next to the CPU isolation field in Expert Settings. In the resulting screen, cut and paste the output here. Based on that, I can suggest an isolation setting to try.

[austinpop]

3 hours ago, Holzohr said:

 

Hmm..  distributed mode? I have tried this (Stylus on the server and StylusEP on the NUC endpoint) but Stylus doesn't "see" the StylusEP. Stylus must support the Squeezebox protocol for this mode I guess? What does work is Stylus plus HQPlayer on the server and Networkaudio activated on the NUC. And it sounds to my ears better than standalone. I have the cheap NUC7CJYH only with the original ps. What will happen if I buy a NUC7i7DNBE (and a Keces P8)? I wonder.

 

Maybe I misunderstood. Hopefully Željko can clarify.

[austinpop]

1 hour ago, RickyV said:

 

 

Sorry I was wondering the same thing for my nuc7i7dn... Since a few days I am able to switch turbo on, hyper threading was already on. This is my output 

 

 

Please specify your use case. Are you running this as a StylusEP endpoint? Or is it a standalone server running Stylus? The process "stylus" runs both the music server and the overall web UI, so you see it running even in an endpoint machine. If a music server, then you want to give stylus some dedicated CPUs. If an endpoint, you can lump it with the default.

 

You could try and compare these (remember each physical core has 2 logical CPUs, i.e. HW threads):

• 0-3 gstp 4-7 < -- this may be good for endpoint (physical core 0-1 for default, core 2-3 for gstp, i.e. the player part of Stylus
• 0-1 stylus 2-3 gstp 4-7 < -- this may be good for standalone  (core 0 for default, core 1 for stylus, core 2-3 for gstp)

 

Beyond this, feel free to experiment.

 

Quote

I have also been playing with the CPU frequency but there seems to be only two settings and that is 1.9ghz and max, which is 3.8~4.2ghz. If I put 1900000 in the max freq field I get 1.9ghz. If I put 1900001 in I get max, 3.8~4.2ghz. Is this correct?

This relates to the issue in the NUC7i7 BIOS with the existing kernel. On the NUC7i7, the max CPU setting does not seem to "take" beyond the base frequency. Whenever Željko updates the kernel, this will then get fixed.

[austinpop]

4 hours ago, Holzohr said:

 

Hmm..  distributed mode? I have tried this (Stylus on the server and StylusEP on the NUC endpoint) but Stylus doesn't "see" the StylusEP. Stylus must support the Squeezebox protocol for this mode I guess? What does work is Stylus plus HQPlayer on the server and Networkaudio activated on the NUC. And it sounds to my ears better than standalone. I have the cheap NUC7CJYH only with the original ps. What will happen if I buy a NUC7i7DNBE (and a Keces P8)? I wonder.

 

If you run Stylus on your server machine, and StylusEP on your DAC-attached endpoint, you should just automagically see the DAC appear as a device in the Settings of your Stylus machine. If you don't, open a ticket. It should be working...

[austinpop]

3 hours ago, Chopin75 said:

What is "automagically", really that magical? haha. 

Now, another dumb question, why would one bother to run and also how can one run Stylus EP while also already using Stylus at server machine?  Actually what exactly is Stylus Endpoint?  The stylus starts to play the music but then instead of going to DAC it is routed to another machine before going to the DAC ?

 

32 minutes ago, HeeBroG said:

 

I believe this scenario would ultimately require x2 Euponhy licences.

I'm not sure if I want to use the distributed method long term particularly if Zeljko doesn't think it is "necessary" given the underlying design principles of Euphony.

When my PH SR7 returns from repair I will be able to run Euphony Stylus on a standalone NUC/Macmini.

Has there been any trials to indicate the distributed mode is better?

 

G

 

 

Whether or not one runs a standalone or distributed (server+endpoint) topology is a matter of choice. There are pros and cons to doing so, and it's up to people to decide if that's what they want to do.

 

The point under discussion here is whether the software enables you to do either. Most music software does. Examples: Roon Core + Bridge, LMS + Squeezelite, HQPlayer + NAA.

Stylus was originally designed to be monolithic, and it was until recently only an option on a standalone machine. The repackaging of Stylus occurred in two steps. People like me  who liked the SQ of Stylus, but were addicted to the UX (user experience) of Roon, were asking Željko if there was a way to make Stylus as a Roon endpoint. This led him to repackage the player portion of Stylus - i.e. StylusEP - and make it accessible using the SlimProto protocol used by LMS and Squeezelite. Having done so, he took the next logical step of enabling remote endpoints on the server piece of Stylus.

 

As regards licenses - yes, if you want to run a server+endpoint, then the server needs to run with the Euphony Server license, and the endpoint with the endpoint license (cheaper).

 

[austinpop]

1 hour ago, HeeBroG said:

 

Thanks Rajiv.

 

I am now trying StylusEP.

 

As for the Squeezelite buffers, the suggested settings are o longer "sticking".

I have 8Gb RAM and what used to be 79% usage is now 4%.

 

Squeezelite buffers are only used when you run Squeezelite, not with StylusEP. The latter manages its own memory, so that's why the memory utilization is much lower.

 

1 hour ago, HeeBroG said:

Here is the screen shot of the CPU isolation as well.

 

Cheers,

 

G

 

Since you're running a NUC7PJYH, which has the Pentium J5005 processor, that does not have hyperthreading, you're limited to 4 CPUs. You could try:

• 0 gstp 1-3 (CPU 0 for default, CPU 1-3 for StylusEP), or
• 0-1 gstp 2-3 (CPU 0-1 for default, CPU 2-3 for StylusEP)

Play with these and other allocations and see what you think.

[austinpop]

5 hours ago, HeeBroG said:

 

Hi Rajiv,

 

Do I recall correctly reading that you use the distributed setup when "grazing" but Stylus monolithic when listening seriously?

This implies Stylus is better in monolithic setup; at least in your system?

 

G

 

I'm not sure to what you're referring when you say "distributed" vs. "monolithic" setup.

 

Let me clarify. My current music server HW is standalone, i.e. I use a single DAC-attached machine. So my point about grazing had only to do with software. When I'm grazing, I use "Roon+StylusEP," running on the one machine. For serious listening, I switch over to Stylus, which also runs on one machine.

 

If by distributed you're referring to Stylus on the server and StylusEP on the endpoint, I don't use that mode obviously, since that requires 2 machines. Stated differently, on a single server, there is no such thing as "Stylus+StylusEP."

 

Is that what you're asking, or did I misunderstand your question?

[austinpop]

6 hours ago, flkin said:

I've been reading this entire thread but can't quite recall if anyone is playing Roon and Stylus EP on separate cores of the same CPU? 

 

Is this possible? On AudioLinux, RoonServer and RoonBridge can't be run together in the same server on different cores as far as I know.

 

Sure. Here is an example. This is on a 6 physical core/12 logical core machine, with HT enabled:

• 0-3 RoonServer 4-7 gstp 8-11 
• This allocates:
  • logical cores 0 to 3 (physical cores 0-1) to default (i.e. everything else)
  • logical cores 4-7 to Roon (strictly: all proccesses whose name contains (see below for an example) the string "RoonServer" in the fully qualified path)
    • 2103 /opt/RoonServer/RoonMono/bin/RoonServer : 4-7
    • 2114 /opt/RoonServer/RoonMono/bin/RoonAppliance : 4-7
    • 2115 /opt/RoonServer/Server/processreaper : 4-7
    • 2188 /opt/RoonServer/RoonMono/bin/RAATServer : 4-7
  • logical cores 8-11 to StylusEP (strictly: all proccesses whose executables match the string "gstp" in the fully qualified path)

I always ensure I allocate all the HW threads in a core to the same workload, to avoid L1/L2 cache contamination

[austinpop]

2 minutes ago, RickyV said:

 

Thanks Rajiv, I copied the above in the core isolation field and it works, I got this in the output screen.

 

i will play around some more this evening. Funny thing is that the temperature of physical core 0 and 1 are about 10 degrees cooler than the other two.

 

If you don’t mind me asking, could you explain the logic behind 0-1 stylus 2-3 gstp 4-7. What is gstp for instance? What does logic core 4-7 do? And you wrote (core 0 for default, core 1 for stylus, core 2-3 for gstp), so why core 2 and 3 for gstp while core 0 is for default - and only core 1 for stylus? I would say 0-1 are for stylus. I am trying to understand.

 

 

 

Please see my post #493 just above yours, where I gave some more detail.

 

It helps to get some terminology straight. On the HW side, a processor chip comprises:

1. Cores: From Intel's website: "Cores is a hardware term that describes the number of independent central processing units in a single computing component (die or chip)." The i7-8650U chip in the NUC7i7DNBE, for example, has 4 cores.
2. HW Threads (hyperthreading): each core can have one or more HW threads, which are independent execution units, but they share some resources on the core, like L1 and L2 caches. The i7-8650U has 2 HW threads/core, or a total of 8 HW threads.

On the OS side, it sees logical CPUs, as presented to it by the BIOS. When you set your BIOS to Hyperthreading (HT) ON, it presents 8 HW threads to the OS, which sees 8 logical CPUs. With HT OFF, it only sees 4.

 

So since you booted with HT ON, Euphony OS sees logical CPUs 0-7, which correspond to physical cores 0-3.

 

I hope this, along with the previous post, help with your understanding?

[austinpop]

On 8/23/2019 at 10:14 PM, HeeBroG said:

At the moment I run AL on my Macmini server and Euphony on my NUC7PJYH.

 

Hi Geoff,

 

The above are OSes. What music software do you run on:

1. the server
2. the NUC endpoint?

 

On 8/23/2019 at 10:14 PM, HeeBroG said:

I tried Stylus EP but found long pauses between tracks and new tracks take 10-20sec to load depending on file size.

 

This isn't my experience at all with StylusEP. Are you loading music files from a very slow device - like a USB stick or a USB 2.0 drive? Also, what network speed are you running? (1Gbps, or 100Mbps, or 10Mbps)?

 

On 8/23/2019 at 10:14 PM, HeeBroG said:

I noticed that memory usage was only 4% and it is not adjustable.

 

Yes, StylusEP does not preallocate large buffers like squeezelite, that I've used extensively, as you know. That notwithstanding, to my ears StylusEP sounds better than SL with large buffers, without the annoying glitchiness that plagues the SL configuration.

 

On 8/23/2019 at 10:14 PM, HeeBroG said:

Things run more smoothly using Stylus Squeezlite and the buffers are adjustable and it so happens that 79% of 8Gb is used with my settings.

 

From this I surmised that Stylus EP doesn't use the RAM in my NUC and is more reliant on AL/Macmini to preload files which I am not sure I can adjust.

 

Could be wrong.

 

In any case in my system there is not much difference in performance.

 

G

 

Ultimately, it's what you hear in your system that counts, so fair enough. But I'm puzzled by the behavior you mentioned, as it's not my experience at all, and suggests you may have some configuration issues.

[austinpop]

14 hours ago, Anwar said:

I had the same experience.  It's like Euphony is trying to produce more details than necessary.  Based on my observation, "Buffer before Play" may not work properly with RAM root option - I can see the HDD light blinking most of the time.

 

Are you using Stylus? "Buffer before play = 100%" and "Use cache" are only applicable to the Stylus server.

[austinpop]

2 hours ago, davide256 said:

how  does one convert the Optane memory to an attachable disk for format/write?

 

Hi David,

 

Optane storage is not a "menory stick" in that it does not plug into a memory slot, nor does the BIOS present it to the OS as RAM.

 

Optane storage (using a technology called 3-D X-Point), while faster than conventional NAND SSD storage, still presents to the OS as an SSD. The HW packaging is in the same M.2 form factor as NVMe SSDs, and so to the OS it just looks like an NVMe SSD, albeit with different speed and latency characteristics.

 

What often confuses matters is that one of the use cases Intel touts for Optane is as a cache accelerator for a slower (but much larger) HDD. This use case requires additional software support (Intel® Rapid Storage Technology), which as far as I know, is only supported on Windows.

 

None of us are using this use case, so for the purposes of this discussion, think of Optane as just an SSD with outstandingly low latency, and very low electrical noise.

[austinpop]

1 hour ago, Anwar said:

I am using Stylus and play from local USB music SSD.  "Buffer before play = 100%" always checked and "use cache" always unchecked.

 

Let me clarify after additional observations.  Apparently for all versions, RAM root or not, for the first folder I "Add & Play" after boot, a green box appears with "Playing dir ....  from song number:1".  When changing tracks, the orange small box "Buffering" never shows up.

 

The second folder I "Add & Play" after boot, the green box appears with 

Buffering
Playing dir ... from song number:1

 

When changing tracks, the orange small box "Buffering" shows up.

 

So it seems there is no buffering for the first folder we selected and played.  I hope others can share their observation too.

 

Interesting. It would be interesting to get some information from Euphony on how to confirm buffering is working.

[austinpop]

1 hour ago, davide256 said:

Is there a way to use Euphony Stylus server to feed a separate Stylus endpoint?

 

Yes.  You run "Stylus" on the server machine, and "StylusEP" on the endpoint machine.

 

Caveat: I have no direct experience with this, but look back a few posts - someone just reported success with this configuration.

[austinpop]

3 hours ago, TheAttorney said:

 

Firstly, the Core 0 temperature increases after applying my particular Isolation settings (as previously posted). After reboot, the temperature reduces to the default values. If I hit the Apply button at this point, the temperature starts increasing again.

 

Secondly, I believe that hitting the Apply button with an empty field will show the actual core values without changing them. If I do this after applying Isolation, the display shows the correct values. If I do this after reboot, the displays shows the default values (0-7 for everything). Tip: Remember to Ctrl/C your isolation settings before clearing the field, so you can quickly re-instate then later.

 

Thirdly, by listening. If I re-hit the Isolation Apply button after reboot, I get a boost in SQ. Any subjective test is not totally reliable, but that's what I hear. It's harder to notice this on reboot because of the larger time lapse and because even default sounds pretty good anyway.

 

 

Are you by any chance setting the isolation flags:

• while booted in ramroot, and
• do you have the flag set to copy app data to root?

If so, you may just be losing the settings on reboot because they were not save back to the disk.

 

I do not have any issue with any of the Expert Settings getting lost on reboot, including the isolation flags.

[austinpop]

2 hours ago, dctom said:

I  asked this earlier, is it possible to use my NAS to play music with Euphony and if it is, how to do it?

At the moment Euphony "sees" the NAS but will not connect. 

 

Would be grateful for any help.

 

My music lives on my NAS, and I use Euphony OS to access it with no problem. You didn't say what music player you're referring to. The process is different depending on the server.

 

Here's my example. The full path to the music folder is \\192.168.0.240\root\My Music\Lossless. Here is what it looks like in Stylus, and in Roon.

 

For Stylus:

 

In Roon:

[austinpop]

@TheAttorney and @RickyV

 

I have seen neither the asymmetric core temps, nor the disappearing core isolation settings you guys are seeing. I'm guessing this is a peculiarity of the NUC7i7DN BIOS.

 

Hope it can be sorted out soon.

[austinpop]

3 hours ago, dctom said:

 

 

Thank you for your response austinpop.

I can run euphony with HQP on my server and send files to my NUC running NAA, HQP picks up the NAS ok. However when I try stylus it wont connect to the NAS, it  shows it as hidden and when I browse network it does not show as available?

 

I don't have roon maybe that is the issue.

 

I have a guess as to what is happening. If you've already mounted a NAS folder through another music app like HQPlayer, it stays mounted when you switch to Stylus. The code to add a "Network folder" in the Library tab seems to not handle a folder that is already mounted by another app. I now remember I ran into this early on, and Željko had to remote in to my system to fix it.

 

One possible solution to fix it yourself is to suppress the mounting of the NAS by HQPlayer, by doing this sequence:

1. Switch to some other server - like Airplay Audio Server
2. Reboot
3. After reboot, you should come back up with Airplay Audio Server running. No NAS folder should be mounted. If you're up for it, you can ssh into the system using credentials audiouser/euphony, and running "df." You shouldn't see any NAS mount in the list. If that sounds too daunting, skip and proceed to step 4.
4. Switch to Stylus
5. Now try adding the NAS via the Library tab.

If that doesn't work, then you should open a support ticket.