You did regarding the NUC platform, I was hoping to squeeze a bit more out of it considering the recent cost. I’m installing a new M.2 NVME 1TB Samsung 870 Plus SSD. I’ll try running Roon on Win32 then if that doesn’t work add a DAS with my external music collection. If that all fails I’ll then try an i9 platform running Roon server.
Depending on how long that takes, you might also look at the newer AMD threadripper CPUs and motherboards that utilize the new PCIe 4.0 specifications which will double the NVMe bandwidth.
DAS setup via USB 3 to i7 NUC, Raid0 array, copying collection should complete tonight. I’ll keep you posted!
Can you please confirm whether this configuration supports a Thunderbolt 3 connected DAS?
Regards,
Zera
I can confirm that the IcyBox USB3 external DAS is too slow @ 150 MB/Sec read/write speed.
I am now trying a Thunderbolt 3 DAS Raid-0 with RoonServer with Win10Pro i7 NUC. The DAS read speed is just under 1GB/s but Roon seems slow in performing audio analysis at approx 1-2 tracks/sec with 5 cores assigned with still over 1.2 million tracks to go!
I also tried running a RoonServer on my W10 gaming PC with a i9-18 core CPU, unfortunately this doesn’t have a T3 port so I could only add the slower IcyBox DAS. This was running faster and smother but hit a limit on the DAS bandwidth.
I really do not want to run a high end Intel or AMD RoonServer due to cost & power considerations, but currently with Roon I see no other options.
Zera
No Thunderbolt included.
ROCK and Nucleus do not support thunderbolt connections unfortunately as Roon does not have linux based drivers. It probably would work with Windows on a NUC tho, but I couldn’t say with 100% certainty.
It does work on the NUC with W10Pro.
This is a contradiction in terms.
4 Likes
Sorry typo as in high-cost, as the quest to get Roon working reliably for with a large music collection has already cost a substantial amount.
Regards,
Zera
But not under Linux - and this is the point. Roon OS is based on Linux.
That’s what I understood, but given you’ve acquired over a million tracks, i.e around 83,000 albums at 12 tracks per album I’m surprised cost is an issue.
Hi Geoff
This is the NUC spec and it does support T3
But not under Linux - and this is the point. Roon OS is based on Linux
Yes I realise that now but my previous post was confirming that the V8 i7 NUC hardware does support a T3 port.
Regards,
Zera
That’s what I understood, but given you’ve acquired over a million tracks, i.e around 83,000 albums at 12 tracks per album I’m surprised cost is an issue.
Yes I can do the maths but this doesn’t really help me solve a technical problem using Roon with a large music collection apart from throwing another 2-3k plus running costs on a top spec Roon server.
Cost is always an issue…
Regards,
Zera
Sequential transfer speed from your disk is unlikely to be a good performance predictor for music analysis: I might be wrong, but I very much doubt that Roon is reading all of the audio data to analyze a track. It definitely needs to scan directories (file names are useful metadata), read the header of each file, and perhaps a small sample of music. Thus, you’re doing essentially a lot of random reads, so IOPS is probably more indicative than sequential transfer rate. The real reason to prefer DAS to NAS in this application (speaking performance-wise) is reduced latency, as that’s the killer. Much lower latency (and 100x more IOPS) is also the driver for SSD performance. I assume your music collection to too large to economically store on SSD, which is currently about $0.10-0.13 per GB, or $400-$500 per 4TB.
Which brings us to the next part of the equation: how long does it take to do a metadata lookup across the Internet? I’m going to make a somewhat educated guess of hundreds of milliseconds on average; it depends where you are, where Roon’s nearest servers are, network topology, etc. (Again, it’s latency that matters more than bandwidth; latency is the harder problem to solve in general for computer systems.) And the 99th percentile time will be in seconds, which you’re likely to hit 10,000 times or so, which will cost you a good fraction of a day.
What this means is that you want many lookups going on in parallel. This means more cores will help. [Yes, I’m somewhat oversimplifying here.] The other thing to do is reduce latency, either through SSDs to store your music, a business level Internet connection with fiber to your home, not oversubscribed (multiply your current ISP bill by somewhere between 10-100), or temporarily moving your equipment to be physically and network-topologically near Roon’s metadata servers.
At 1 track per second, a million tracks will take about two weeks. Go take a nice vacation, see some live music, and when you get home Roon will be done. This is likely to have the best ROI for you.
Good luck!
1 Like
It does analyse the whole track in order to report on the dynamic range and generate the waveform graphic.
That said, CPU performance is the single largest factor when performing this audio analysis. Disk I/O (of the music storage) is somewhat trivial in comparison.
Note this audio analysis is a separate task to track identification, metadata pull down and “clumping” to form albums.
I totally concur, I have that very discussion with my client’s all the time when they are investigating moving from onsite data centres to remote (aka cloud) based hosting.
1 Like
With all due respect, you don’t have a technical problem.
3 Likes
1-2s per track is absolutely fine. 10-15s per album. The issue is the multiplier by number of tracks. It is an extreme collection.
It would not be unreasonable to just let this run for a week and get the job done. I have done this many times (as one of the earliest alpha testers it was a frequent task) and often it would run for many days to redo the analysis.
Once it is done then backup and get on with the music.
2 Likes
This raises a question that I have wondered about: For music that comes from Qobuz/Tidal, presumably there is no analysis. So: Does lack of analysis mean music from such sources is “less good,” sounds different, in comparison with music on site that has been analyzed?