Our fastest server EVER - Serviceable Whonnock

[James]

Buy a SABRENT Rocket 4 Plus 8TB NVME M.2 SSD: https://lmg.gg/qgDC3
Buy an ICY DOCK EZCONVERT MB705M2P-B NVMe SSD to 2.5" U.2 SSD Converter: https://lmg.gg/fjTFz
Buy a StarTech U2M2E125 NVMe SSD to 2.5" U.2 SSD Converter: https://lmg.gg/76FvY
Buy a GIGABYTE R282-Z9G 2U Rackmount Server: https://lmg.gg/CxlsQ
Buy an AMD EPYC Milan 75F3 CPU: https://lmg.gg/DzxJF
Buy Micron 3200MHz CL22 DDR4 2x4GB ECC Memory: https://lmg.gg/k1myU
Buy an Ableconn E1s-DT157 NVMe M.2 SSD to NVMe EDSFF E1.S SSD Adapter: https://lmg.gg/PC1kk

Purchases made through some store links may provide some compensation to Linus Media Group.

 

We made a big mistake when setting up our main editing server Whonnock - it wasn’t serviceable. Today we fix that with a brand new machine and brand new sabrent nvme drives!

 

 

 

[AbydosOne]

4 minutes ago, James said:

EEC Memory

*ECC

[Hyperthermia]

cant believe they made linus sex tips real

[sageofredondo]

I am thinking of a home all SSD server with one (or two) of those ASUS 4x nvme pcie cards with consumer ssds. One common complain I hear is that all the ssds could fail at the same time since they have write maximums. That did not happen in this case in the video; I am inferring from the video that you ran the server for months until you replaced it with the new server so they did not fail at the same time. Is there any truth to it in practice?

[NF-A12x25]

22 minutes ago, Hyperthermia said:

cant believe they made linus sex tips real

fact

[leadeater]

Damn those converters are really nice, super cool for home labs and used servers.

[Needfuldoer]

5 hours ago, sageofredondo said:

I am thinking of a home all SSD server with one (or two) of those ASUS 4x nvme pcie cards with consumer ssds. One common complain I hear is that all the ssds could fail at the same time since they have write maximums. That did not happen in this case in the video; I am inferring from the video that you ran the server for months until you replaced it with the new server so they did not fail at the same time. Is there any truth to it in practice?

We can help you with this down in the server/NAS board. 

 

However, an all-NVMe NAS is completely pointless for a home application. (Especially if all your networking gear is just regular Gigabit Ethernet, where all your transfers will be limited to about 100 MB/sec.) By all means have an SSD for your Plex/Jellyfin/Subsonic library metadata, but hard drives still make more sense for bulk storage in that application.

[Nystemy]

2 hours ago, sageofredondo said:

I am thinking of a home all SSD server with one (or two) of those ASUS 4x nvme pcie cards with consumer ssds. One common complain I hear is that all the ssds could fail at the same time since they have write maximums. That did not happen in this case in the video; I am inferring from the video that you ran the server for months until you replaced it with the new server so they did not fail at the same time. Is there any truth to it in practice?

The failure mode of SSDs is a bit complex in practice.

 

Effectively speaking, Flash is inherently forgetful. The data is just stored as a bit of charge, it does slowly leak over time and eventually the Flash chip doesn't know what it originally were. Manufacturing tolerances also adds the fact that different chips forgets at different rates. Beyond that there is also the fact that with each write/erase cycle to a cell, it leaks faster. Eventually it forgets its contents fast enough to not be practical as data storage for a given application, ie data retention is quite limited in comparison to magnetic storage. _{(scratch disk applications can usually use a worn out SSD just fine for quite some time.)}

Some people will pull out Microchip's data retention specs for their industrial flash chips and microcontrollers, stating that the "data retention of flash is 20 years at 85 C after 20K writes." But that is like saying that any car can drive at 700+ km/h just because a land speed racer did do that. _{(Industrial grade flash isn't remotely comparable to enterprise grade flash. Industrial flash is built for applications requiring long term reliability and where it is allowed to cost more than a dollar per MB to achieve said reliability in harsh environments, while Enterprise grade flash is built for the lowest possible GB/$ for use in mass storage in fairly well maintained ambient environments.)}

But in general, flash doesn't really just give up the ghost after X hours unlike the typical hard drive that often is more predictable.
However, there is still some single points of failure in these devices, a cracked MLC capacitor shorting out the power rail will often make the drive into a brick, same for a few cracked solder joints, an excessively bent PCB developing tears/cracks in the internal copper tracks or just good old ESD, tin whiskers, conductive gunk/corrosion, or a slew of other issues. _{(At least SSDs don't also have the mechanical wear of HDDs that makes the list of single point of failures far longer...)}

An SSD is fairly random when it will die/stop-working. But it can tend to have corrupted data in it long before it dies if one don't use it a lot.
HDDs are a bit more predictable due to their mechanical nature, but even here a given drive can fail more or less whenever, some last a few weeks, others go 10x beyond Mean Time Between Failures. But for thousands of drives, it will on average be quite close to MTBF.

In the end.
If one cares about data retention in flash media, it is quite wise to check the data on a somewhat regular interval and correct any errors that are found. _{(I would say yearly, if not monthly if the drives are getting on the old side of things. Also, this is wise to do for HDDs too, they can also have bit errors induced over time even if this is more rare. This is also not a particularly atypical feature of most storage software/controllers, however one can do like LMG did one time and forget to turn that feature on, and I wouldn't be surprised if some SSD controllers do this in the background.)}

[themrsbusta]

[Somerandomtechyboi]

Why 1rx16 rams

seems kinda dumb to populate all the ram channels but only have 1rx16 rams

[leadeater]

1 hour ago, Somerandomtechyboi said:

Why 1rx16 rams

seems kinda dumb to populate all the ram channels but only have 1rx16 rams

Because ram capacity and maximum possible throughput doesn't really matter in this situation and configuration. Simply having all the ram channels populated is quite satisfactory. 

[sageofredondo]

5 hours ago, Needfuldoer said:

We can help you with this down in the server/NAS board. 

 

However, an all-NVMe NAS is completely pointless for a home application. (Especially if all your networking gear is just regular Gigabit Ethernet, where all your transfers will be limited to about 100 MB/sec.) By all means have an SSD for your Plex/Jellyfin/Subsonic library metadata, but hard drives still make more sense for bulk storage in that application.

 

I am limited in space right now and I plan to move again. I also have sensory issues. The only moving parts being noctua fans is a big plus for me. Not having noise or deal with a NAS card and more resistance to moving damage is important for me.

[dogwitch]

13 minutes ago, sageofredondo said:

 

I am limited in space right now and I plan to move again. I also have sensory issues. The only moving parts being noctua fans is a big plus for me. Not having noise or deal with a NAS card and more resistance to moving damage is important for me.

i find using 200 mm fans size with slower  overall speed but same air volume. much better then tiny fans.

also i have  just the front case fans and gpu led. to to know pc on!!!

its that quite

[sageofredondo]

57 minutes ago, dogwitch said:

i find using 200 mm fans size with slower  overall speed but same air volume. much better then tiny fans.

also i have  just the front case fans and gpu led. to to know pc on!!!

its that quite

Your avatar, a smallish Rainbow Dash is adorable @dogwitch

[dogwitch]

4 hours ago, sageofredondo said:

Your avatar, a smallish Rainbow Dash is adorable @dogwitch

thank you.

i meet the voice actress that voice her.

[NemesisPrime_691]

Sorry, a storage/servers noob here, but don't these convertors from NVMe M.2 to the server sled format create more points of failure? Are these convertors also enterprise level (low failure rate and such)? 

 

And also, this solves the issues of the drive replacement, but checking which drive has failed, is that something their server software can let them know of or they have to keep checking which one failed? Probably the indicator LEDs on top of each drive bay?

[leadeater]

4 hours ago, NemesisPrime_691 said:

Sorry, a storage/servers noob here, but don't these convertors from NVMe M.2 to the server sled format create more points of failure? Are these convertors also enterprise level (low failure rate and such)? 

There isn't any logic to them, it's just copper traces and connectors/physical interfaces. There's no electronics to fail so the added risk is very low. Of course this isn't really something you "should do" but you can and isn't a massive risk in general.

[itsVANGETTI]

When will you guys deploy the Weka FS server & are you guys using the same server?