Computer Storage - Best Practices

[Nenon]

2 hours ago, jabbr said:

5. NAS

 

I am 100% confident that employing a NAS running ZFS with a good RAM buffer, and connected by a good fiberoptic Ethernet switch, that assuming the NAS is reasonably decent, that there will not be SQ effects of the hard drives, nor will the PSU supplying the NAS be audible.

 

Indeed I have compared 10Gbe Intel x520 cards with special Solarflare NICs as well as Connextx-4 100 Gbe Mellanox NICs and these on the NAS "sound" the same. (do you know what the jitter specs on a 100Gbe NIC have to be in order to hit that eye diagram?  )

 

http://forms.solarflare.com/Media/Default/PDFs/Platforms/Solarflare_SFN7322F_10GbE_Adapter_Brief.pdf

https://www.mellanox.com/related-docs/prod_adapter_cards/PB_ConnectX-4_EN_Card.pdf

https://www.intel.com/content/www/us/en/ethernet-products/converged-network-adapters/ethernet-x520-server-adapters-brief.html

 

http://uk.tek.com/dl/65W_28494_1_LR_Letter.pdf

https://www.evaluationengineering.com/test-issues-techniques/article/13007997/100-gbs-physicallayer-testing-tips-and-tricks

 

It seems like a lot of people agree with that here. Especially those who use JCAT network cards and/or audiophile switches. I think that option goes to the top of the list and might be the best (sounding) option for large music collections. 

 

@jabbr - any recommendations for a decent NAS? 

I believe @austinpop uses a 4 bay Synology NAS.

 

From what I see so far, people tend to prefer (in that order):

1. NAS. JCAT Net / audiophile switches make that option even better. 

2 - a. An external USB HDD with external power supply. Adding an external USB card could be an improvement. 

2 - b. USB Tower / SDXC cards / LPS

3. HDD

4. SSD

 

Of course there are as many opinions as people, so not to start a debate, especially knowing that a few people think SSD is better than HDD. But that seems to be the general consensus so far. I did not add Optane as it's not available in Terabyte sizes yet. 

[jabbr]

23 minutes ago, Nenon said:

@jabbr - any recommendations for a decent NAS? 

I believe @austinpop uses a 4 bay Synology NAS.

 

If you want something off the shelf I suppose Synology is good. Given the interest in tweaking here, using audiolinux etc, I advocate getting a server (which doesn't need a lot of CPU power) which has 8-12 bays and a SAS 6 or 12G controller. For example I just picked up a Dell T420 for about $500 on ebay. I've also built a low power Xeon server which is still running strong with an LSI SAS3 controller and an Intel x520 NIC. 

 

Run Ubuntu server which comes with the ZFS filesystem standard. The more RAM you have the more it caches to memory. I have 160 Gb RAM, so *alot*.

 

I don't use audiophile NICs or audiophile switches because frankly mine are much much better. Regardless if you use a single good fiberoptic segment the upstream stuff doesn't really matter. I say this but I have professional grade network equipment in my house -- just because its fun --- there is absolutely no need to use 100Gbe for audio!

 

Right now I'm typing this, drinking some excellent Rioja, and listening to DSD512 streamed by wifi to my NAA->DAC. The bits are stored on a NAS which is accessed by the upsampling server (HQPe) and then sent up to the Wifi via fiber ... those hard drives are so far in the background there is no SQ.

 

But in all seriousness, here is the computer storage best practice: use ZFS and build your pools from mirrors. I add stripes of mirrors when I want to expand a pool. I also have multiple pools. Mirror!

[Nenon]

3 minutes ago, jabbr said:

I don't use audiophile NICs or audiophile switches because frankly mine are much much better.

 

Solarflare and Nexus? 

[jabbr]

3 minutes ago, Nenon said:

 

Solarflare and Nexus? 

The Intel x520 cards are great -- all the 10Gbe cards have low jitter -- the Solarflare that I referenced have crazy low latency and designed for transaction processing/trading shops, which have now moved on to 100Gbe (200Gbe coming out as we speak, and 400Gbe on the way) ... needless to say the jitter window at 100 Gbe is nonexistant. Mellanox is cleaning up here. The articles I referenced discuss this. 

 

What I am telling everyone is that I can't hear a difference even with my special SOA cards compared to Intel x520 so get Solarflare "Flareon Ultra" if you care, and by all means swap out different SFP(+) modules but its all good.

[Ralf11]

some excellent Rioja ---> Gran Reserva??

 

 

somebody should buy up some fiber equipment; stamp Audiophile on it; add Bubinga wood panels and go to town...

[BigAlMc]

I have a 1tb m.2 nvme SSD in my NUCi7DNKE server and whilst its sounding great to my ears I'm wondering if there's an improvement to be had in removing it.

 

The SSD is only storing the library and Roon DB backups as I'm running Audiolinux in RAM.

 

Am toying with two options.

 

A. An external m.2 nvme enclosure powered by LPS-1 (as I have a spare one I can use).

 

This is my preferred plan as the money spent on the SSD isn't wasted and I can simply move it.

 

B. Move the library to an external SD card hub or USB stick hub and again power that with the spare LPS-1.

 

Can anyone recommend an external m.2 nvme enclosure that accepts external DC power?

 

Ditto for the USB hub?

 

And will these work with the 9v output of an LPS-1?

 

Many Thanks,

Alan

 

[R1200CL]

On 3/27/2019 at 8:39 AM, romaz said:

With the JCAT card, I am finding that the media storage device no longer matters.  At all.

 

This sounds very promising for the upcoming Uptone Audio EtherRegen 😀

 

As I expect it to be equal or better than the JCAT.

 

[romaz]

32 minutes ago, R1200CL said:

 

This sounds very promising for the upcoming Uptone Audio EtherRegen 😀

 

As I expect it to be equal or better than the JCAT.

 

 

This is very true.  The one possible advantage of the JCAT card is that while you are applying a cleanly reclocked signal to your music server in the same way that the EtherREGEN would, the JCAT card is also replacing (or removing from your signal path) the dirty Ethernet port that is in your music server.  Even with SOtM's sNH-10G switch in place, I found the JCAT card to still make a significant difference.

[R1200CL]

5 minutes ago, romaz said:

the JCAT card is also replacing (or removing from your signal path) the dirty Ethernet port that is in your music server.

 

So a JCAT on a SonicTransporter or a NAS would be something 😀

 

But not a possible solution.....

 

Not even on a NUC ?

[darkfrank]

16 hours ago, romaz said:

 

This is very true.  The one possible advantage of the JCAT card is that while you are applying a cleanly reclocked signal to your music server in the same way that the EtherREGEN would, the JCAT card is also replacing (or removing from your signal path) the dirty Ethernet port that is in your music server.  Even with SOtM's sNH-10G switch in place, I found the JCAT card to still make a significant difference.

Hello Romaz, can you share your network topology between the music server and the NAS?

 

What kind of switches are being used? Dual switches with fiber isolation? 

[numlog]

In regards to SSD vs HDD,  I was reworking the external supplies for my PCIe SATA/USB cards and drive, adding pre-regulators before primary regulators, either with similar ultra low noise types (lt3045, TPS7A4700) or standard (LM350) .

 

Previously when used with the SSD, the drive supply's regulator would run cool (except during writing), only with HDD would it run warm/hot during playback. This was obviously bad yet HDD still sounded best, balancing the load with a pre-regulator should make it even better... right?

Well apparently not, with pre-regulators added to drive, USB and SATA supplies the SSD had a leap in SQ while the HDD is only slightly better.

Most surprising is that the SSD harshness has mostly been lifted, the HDD is still ''smoother'' but with the slight muddiness it adds its hard to go back to it.

 

So in short I guess HDDz are very forgiving of the rest of the system while SSDs are the opposite but they ''scale'' exceptionally well.

They must be highly sensitivity to power supply ripple and noise, because when previously testing a non-optane Kingston NVMe M.2 Drive (i.e noisy MoBo supply) it was even worse than the SATA SSD without the recent power supply improvements.

 

Also the 16GB M.2 optane card I recently got managed to sound better than the SATA SSD before the PS changes so I can only imagine how good the Optane might sound with it... of course even it was possible its too small for storage

[sandyk]

1 hour ago, numlog said:

with pre-regulators added to drive, USB and SATA supplies the SSD had a leap in SQ while the HDD is only slightly better.

Most surprising is that the SSD harshness has mostly been lifted, the HDD is still ''smoother'' but with the slight muddiness it adds its hard to go back to it.

 

 That is exactly what I have been saying with my method of pre-regulating the +12V supply down to +5V followed by a very low noise (<4uV ) JLH PSU add-on. You really need to do the same for the OS SSD as well as the Music SSD though.

 It's harder to do this with the original smaller LT3045s though with their maximum Current Capability of 500MA , although you could use 2 lots of 2 in parallel, but all this could become unwieldy.

[numlog]

1 hour ago, sandyk said:

 

 That is exactly what I have been saying with my method of pre-regulating the +12V supply down to +5V followed by a very low noise (<4uV ) JLH PSU add-on. You really need to do the same for the OS SSD as well as the Music SSD though.

 It's harder to do this with the original smaller LT3045s though with their maximum Current Capability of 500MA , although you could use 2 lots of 2 in parallel, but all this could become unwieldy.

Your suggestion for the power supply was excellent, just for unexpected reasons, I didnt expect the SSD to benefit so much, and HDD so little, for the reasons you explained.

Your reason of using pre-reg is to use 12V ATX supply and avoid possible ground loops of external supply, my external supply is 8V from rectifier and with very short distance to transformer and no earthing, so 1 reg should be ok for SSD if heat is ok?

Also some effort was made to remove low ESR caps (tantalum) from reg outputs as you suggested, which helped but didnt fix the problem.

 

The combination of pre-reg and JLH add-on must be an exceptionally high quality power source compared to most, could that be the most significant factor in SSD sound quality?

 

Perhaps I am oversimplifying the use of the pre-reg also, even if there was no obvious heat problem dropping 3.5V with ULN reg they might not sound very good.

 

 

[sandyk]

53 minutes ago, numlog said:

The combination of pre-reg and JLH add-on must be an exceptionally high quality power source compared to most, could that be the most significant factor in SSD sound quality?

 What you are doing should work quite well. Even doing what I am doing with just the OS SSD gives an improvement.

 If you have both a 1A LM317 based PSU and a 1A LT3045 try the LM317T set to say 2V above the required output voltage then into a 1A LT3045 in a temporary set up. You can fiddle with capacitors later to achieve the best results in your own setup.

[numlog]

LM350 pre-reg were used for PCIe cards where more dissipation was required for 3.3V while 2 TPS7A4700 (1A) in series were used for the drive, roughly 1.7V drop each. 

 

looking at this example of LT3045, Its really surprising how much more effective regulators can be in series, also surprising how 1 reg that should push PSRR into ''inaudible'' levels isnt enough. LT3045 is on -80dB... so I guess it makes perfect sense! 😁

 https://www.diyaudio.com/forums/vendor-s-bazaar/310402-fs-ultra-low-noise-power-supply-lt3045-based-pcb-15.html#post5424495

[sandyk]

5 minutes ago, numlog said:

LM350 pre-reg were used for PCIe cards where more dissipation was required for 3.3V while 2 TPS7A4700 (1A) in series were used for the drive, roughly 1.7V drop each. 

 

looking at this example of LT3045, Its really surprising how much more effective regulators can be in series, also surprising how 1 reg that should push PSRR into ''inaudible'' levels isnt enough.

 https://www.diyaudio.com/forums/vendor-s-bazaar/310402-fs-ultra-low-noise-power-supply-lt3045-based-pcb-15.html#post5424495

Quote

get-it, I think you theories are simplified a lot and needs to be verified. The real result will not be as good as you think.

 

I agree with the well experienced Peranders.

[BigAlMc]

On 3/28/2019 at 6:19 PM, davide256 said:

this is basically a device for photographers who want a fast solution for use of their photo memory cards

 

USB tower

https://www.bhphotovideo.com/c/product/1002489-REG/lexar_lrwhr1rbna_professional_workflow_hub_hr1.html/?ap=y&gclid=Cj0KCQjw4fHkBRDcARIsACV58_EJkeaGjDUmt0HiP1d4nNOvHsdbZf0893E1Bo5qf7cyvkb8arD__n0aAq3yEALw_wcB&lsft=BI%3A514&smp=Y

 

you will need at least 1 SDXC card reader, seems like these aren't in supply at the moment

 

https://www.bhphotovideo.com/c/product/1186727-REG/lexar_lrwsr2tbna_professional_workflow_sr2_sd.html

 

the voltage is standard USB 5v so if the original LPS1 had 5V option, it can be used

 

there are faster cards now but the 95 MBs cards are more attractively priced, possibly closeout pricing ?

 

https://www.bhphotovideo.com/c/product/1186714-REG/lexar_lsd512cbnl633_512gb_professional_uhs_i_sdxc.html

 

Hi @davide256,

 

So the HR1, SR2 and SD card are all here. Planning on comparing to my SSD this weekend but obviously the need to remove an SSD means A-B-ing needs a bit of a gap.

 

But couple questions meantime:

 

Firstly, I noticed the HR1 light stays powered on when I remove the AC power supply. So do I need to disable USB power? And is it possible to do so?

 

Secondly, do you use an audiophile USB cable or short USB attachment in between the HR1 and your server? 

 

Many Thanks,

Alan

[davide256]

12 minutes ago, BigAlMc said:

 

Hi @davide256,

 

So the HR1, SR2 and SD card are all here. Planning on comparing to my SSD this weekend but obviously the need to remove an SSD means A-B-ing needs a bit of a gap.

 

But couple questions meantime:

 

Firstly, I noticed the HR1 light stays powered on when I remove the AC power supply. So do I need to disable USB power? And is it possible to do so?

 

Secondly, do you use an audiophile USB cable or short USB attachment in between the HR1 and your server? 

 

Many Thanks,

Alan

It doesn't appear that there is any way to run off adaptor power only, although we could try the trick of taping over the USB 5v lead and see if all still works.

 

Using a plain vanilla USB3 cable with the device, for some reason the Wireworld starlight USB3 cable I have and it are temperamental about working together,

50/50 chance the device won't be seen at boot.

 

Might be like shooting fish in a barrel to compare an internally powered SATA connected SSD to an external LPS 1.2 powered USB3 connected

SDXC media solution. I plan to order a few external  2.5"  USB3 connected cases that can accept external power so that I can try a like comparison

with SSD... I'm not sold yet that SSD has any problems other than the environment powering it. 

 

 

[numlog]

@sandykHave you experimented with allocation unit size for storage in windows? seems like something you would be interested in.

 

I was testing different partitions with different file system and allocation sizes, there may be differences here but it doesnt seem to very consistent when verified by ear, they are so small that they may not exist or may be result of small drifts in performance  of rest the system (e.g class A HPamp, slight drift in voltage will affect the distortion profile, among many other things, even in the PC alone)

 

Something that might be important for the differences to even exist is buffer sizes, In XXHE you have indepth control over buffer size and latency. In general the absolute minimum value without stutter offered best SQ and going slightly above these minimums tends to lose some special qualities, when testing different partitions it seemed like increasing buffer size beyond absolute min made hearing any differences impossible. e.g Increased difficulty in the comparisons was noted when listening with high res files on the partitions, where the buffer size was increased roughly from the redbook min, instead  fine tuning to the new absolute minimum buffer for that high res sample rate.

I think this is intended purpose use of buffer, right? to remove these differences, but for whatever reason the larger buffer imparts its own influence on the sound, so for absolute best SQ its not a an ideal solution.

 

For partitions, FAT32 has been most ''prefered'' choice in older posts.

 Im assuming at the very least there could be little reason for NTFS to be BETTER for audio, it seemed that way after a short comparison so it was left out to make things simpler.

Comparing exFAT and FAT32 wasnt very conclusive, but FAT32 did seem to have the edge.

Allocation sizes seemed to have the most difference, 512B vs 64K showed patterns that made guessing between them more consistent, but the ''winner'' seem to vary by each track.

Its possible based on your theory that these differences were imparted when the files were copied to both the partitions, and are the result in slight fluactations in load/noise on the CPU and data interfaces rather than partitions differences.

To help minimize those possible differences I used a RAM Disk as buffer for the files before copying to the partitions.

 

With that said,  exFAT is unique in that it allows very large allocation sizes, up to 32MB.

a 32MB partition seems to a more significant effect on the sound, out all the blind comparisons the 32MB partition was chosen most consistently as the best sounding (compared to 64k exFAT and FAT32) ... could have been a lucky streak but a good start.

 

Also an Optance drive was used for this, it is significantly more resolving than either SSD or HDD as storage so useful for the purpose of testing

 

XXHE aswell is running from RAM, performing file conversion on RAM, aswell as being a memory player, all for best SQ. SW that doesnt use RAM may bring out these differences  more clearly... of course this would only be for the purpose of testing as utilizing RAM is far great priority over whats mentioned above.

[sandyk]

41 minutes ago, numlog said:

sandykHave you experimented with allocation unit size for storage in windows? seems like something you would be interested in.

 Hi

 No, I haven't played around in that area so far, as I can obtain much greater differences playing around in the PSU area.

Also, my very best files are saved to USB memory using a Regen powered via a very low noise PSU powered from a 12V Li Ion battery to then a JLH PSU add-on to get away from mains earth related problems. The JLH that I use has a simulated capacitance of >2FARADS which is getting into UltraCap territory.

 USB memory is another hairy area as some later ones are using ExFat and you may need to use a 3rd party application with Windows to change anything. I haven't gone too deeply into it, but I do have a small preference for FAT32, but the PSU area is to me far more revealing at this point in time.. 

52 minutes ago, numlog said:

I think this is intended purpose use of buffer, right? to remove these differences, but for whatever reason the larger buffer imparts its own influence on the sound, so for absolute best SQ its not a an ideal solution.

 

 That's an interesting one, as Gary (one and a half) would like to know more about too, as he reports the same too.

[davide256]

On 4/4/2019 at 5:01 PM, BigAlMc said:

 

Hi @davide256,

 

So the HR1, SR2 and SD card are all here. Planning on comparing to my SSD this weekend but obviously the need to remove an SSD means A-B-ing needs a bit of a gap.

 

But couple questions meantime:

 

Firstly, I noticed the HR1 light stays powered on when I remove the AC power supply. So do I need to disable USB power? And is it possible to do so?

 

Secondly, do you use an audiophile USB cable or short USB attachment in between the HR1 and your server? 

 

Many Thanks,

Alan

Tried  taping off the 5v Vcc pin 1on USB cable, the Lexar USB tower  does require the 5v USB lead be active in order to work. Awaiting another

2.5" 5V externally power-able enclosure to compare SSD, ruled out the 1 I already have because it has a built in USB3 hub... which also annoyingly means its

not usable/recognized for PS4 external storage.

[Poldi]

@romaz I found your comments regarding storage very insightful. Yet, I am in need of a little clarification:

 

You mentioned that the type of media storage doesn't matter at all as long as the track is loaded into RAM before playback (for example Euphony stylus or Audiolinux ram play option with MPD). So in this case there isn't any downside to using a SSD as storage device instead of a HDD and no SSD harshness? That would be nice as a HDD would be too loud in my living room.

 

You also mentioned that with the JCAT Femto Card the quality of the NAS doesn't matter any more in your opinion. As I am not familiar with your setup: Did you by "NAS" mean your music server or do you have a separate NAS just for storage and a separate music server for running the server software? So in the latter case the server would still matter, I suppose.

And if you do have separate music server and NAS is there any sound-related reason for not just using an external HDD/SSD with the server instead of a NAS?

 

[sandyk]

 

Does using a JCat Femto card mean that you no longer need to use an  aftermarket (usually expensive) USB cable such as the Lush  because a generic USB cable of the same length MUST then sound the same  ? 

 

 

[chauphuong]

Regarding ramdisk, do you folks think installing minimserver (or the likes) to ramdisks will impro SQ?

[Dutch]

9 hours ago, sandyk said:

 

Does using a JCat Femto card mean that you no longer need to use an  aftermarket (usually expensive) USB cable such as the Lush  because a generic USB cable of the same length MUST then sound the same  ? 

 

 

 

Just as an FYI; in this case Romaz referred to the JCAT NET card, so the network interface card.