NIC teaming for Windows 10 sudo-server?

[Gerr]

I have a Windows 10 system that I want to turn into a multi-function system.

 

Uses:

Gaming(SteamLink)

Plex server

NVR

NAS

 

Hardware:

i7-4790K @ 4.5Ghz(1.28v)

Asus Z97 Pro Gaming mobo

32GB DDR3@2133 CL9

Asus GTX 1060 6GB

LSI 8-port SATA/SAS HBA

275GB Corsair MX300 M.2 (OS & App drive)

1TB WD Black (game drive)

6TB(4x3TB RAID10) WD Red (Plex media drives)

250GB Samsung 850 EVO (Plex app & metadata drive)

3TB WD Purple (NVR drive)

3TB(2x3TB-RAID1) HGST NAS (redundant NAS storage)

Blu-Ray optical drive

Intel i350 4-port NIC

750W Seasonic Gold PSU

 

I really don't want all of this going over a single 1G network link.  I have a switch that supports LACP so I would like to team all 4 ports on that i350 NIC together.  I am not looking for load balancing to speed up a single connection, but instead want enough bandwidth when multiple devices connect to this PC, including multiple IP cameras & 1-3 Plex clients.

 

From what I have read, the Intel NIC drivers broke NIC teaming in Windows 10, but I also read that you can manually do it inside PowerShell.  This whole build depends on NIC teaming and I would really rather have this single system than multiple systems running 24/7 if possible.

 

If I can't do this in Windows 10, I do have a copy of Windows 8.1 Pro that I could install, but would prefer to stay on Windows 10 if possible.

 

Thoughts?

[Electronics Wizardy]

The intel nic drivers let you setup teaming. 

 

The powershell teaming is disabled in windows 10, and can only be used in server(if your a student you can often get it for free).

[Electronics Wizardy]

16 minutes ago, Gerr said:

6TB(4x3TB RAID10) WD Red (Plex media drives)

Id run this in raid 5, you don't need 10 as you don't need lots of iops.

 

Also if you haven't bought the drives yet, it just make one large raid 5 or 6, instead of lots of smaller arrays, and then partition it out.

 

ALso Id use storage spaces in windows, not mobo raid, unless you have a hardware raid card.

[Gerr]

All drives already bought.

 

Was using mobo RAID, so thought 10 would be better than 5.  Only 5400RPM drives, so speed could become an issue on multiple streams.

 

I like keeping drive roles separate incase of failure as some need redundancy and/or backup and some don't need any.

 

Have not heard good things about Windows Storage Spaces, so not sure I want to use it.

 

I do have a copy of Windows 2016 Server Essentials, but you can't game on that.

[Electronics Wizardy]

58 minutes ago, Gerr said:

All drives already bought.

 

Was using mobo RAID, so thought 10 would be better than 5.  Only 5400RPM drives, so speed could become an issue on multiple streams.

 

I like keeping drive roles separate incase of failure as some need redundancy and/or backup and some don't need any.

 

Have not heard good things about Windows Storage Spaces, so not sure I want to use it.

 

I do have a copy of Windows 2016 Server Essentials, but you can't game on that.

You can game on windows server. It's basically windows 10 with the server dashboard. 

 

Storage spaces on windows(esp in 2016) is much better than motherboard raid. 

 

Riad 5 is normally faster than raid 10 in sequential, which media servers are mainly doing. Raid 10 is only faster with lots of small files. 

[Mikensan]

I might not be right, but I'm pretty sure Windows NIC teaming is not LACP. You could in turn just have each gigabit link assigned a different IP address, and then tell said services to listen/bind to those specific IP addresses. Now a piece of software would have a dedicated gigabit link. 

 

You have 5 gigabit ports so:

 

Port 1: plex / internet / gaming, 192.168.0.2/24 (so that you can easily share out plex to the world wide web if needed)

Port 1of4: NAS 192.168.0.3/24

Port 2of4: NVR 192.168.1.1/24

 

 

Or something like that anyway. NIC teaming / LACP will not divide up the bandwidth per se anyway, so you're still only going to get a gigabit per service/software anywho (unless you had 10+ 1080p cameras then it might be beneficial).

[Dark]

5400rpm drives will serve streams perfectly fine.  They will suffer heavily with random seeks and multiple small reads.

 

 

[Falconevo]

Windows 10 doesn't support LACP, unlike Server 2012 R2/2016 which supports LACP native regardless of the network interface driver..

I believe one of early builds of Windows 10 supported this, they have since removed the feature for unknown reasons.  You can run the PS commands but the LACP interface stays disabled after creation on the newer builds of Windows 10.

Why not just install Server 2012 or 2016 and leave it running non licensed, it sounds more suitable for you?  You can do this although obviously not advocated by Microsoft all the features remain enabled.

[Gerr]

This system would be used as an every day PC and gaming PC as well, neither which Windows Server would do well at.

 

I read that Windows 8.1 still supports LACP via Intel drives, so might just go that route as this is only a secondary gaming PC, so don't need DX12 on it.

[leadeater]

18 minutes ago, Gerr said:

This system would be used as an every day PC and gaming PC as well, neither which Windows Server would do well at.

 

I read that Windows 8.1 still supports LACP via Intel drives, so might just go that route as this is only a secondary gaming PC, so don't need DX12 on it.

Other than cost there is nothing stopping you from using Windows Server on a workstation. Fundamentally they are the same and will not have any impact on game play, I have run server operating systems on my gaming desktop before. There are even dedicated communities that give guides on how to fully convert a Windows Server OS for workstation use.

 

Not that I think you should run a server OS or even need LACP. What @Mikensan suggested will do what you need, put your two NICs on different subnets and access the server/services you want on the subnet for the NIC you want it to use. You don't need any switch support for this. For the other devices that only have 1 NIC give each of those devices two IP addresses, one on the router IP range and one for the dedicated NIC subnet.

[Mikensan]

Using Windows Server for your OS is fine, you just install Desktop Services feature to get a majority of the frills. Disable enhanced security for internet explorer, download chrome - and you're well on your way to using Server 2016 I mean uh Windows 10 ;-).

 

You don't really gain anything by using Windows Server unless you plan to actually use one of the roles and features it offers.

[Falconevo]

Server 2016 has the same core build as Windows 10, can be enabled to support all of the same features + more.

I would go for Windows 2016 using Windows 10 drivers for your GPU and any other additional peripheries, then you can take advantage of inbuilt teaming within server manager.  Not many changes are required to turn 2016 in to a usable desktop operating system far more features available.

DX12 is present in 2016 by the way and it will be updated more often in relation to security fixes.  I have used server operating systems as daily drivers including gaming for a long time, very rarely have any issues.

 

[Blake]

On 12/10/2016 at 5:20 AM, Electronics Wizardy said:

Id run this in raid 5, you don't need 10 as you don't need lots of iops.

 

Also if you haven't bought the drives yet, it just make one large raid 5 or 6, instead of lots of smaller arrays, and then partition it out.

 

ALso Id use storage spaces in windows, not mobo raid, unless you have a hardware raid card.

No, with RAID 5 you have like a 75% chance of the array failing before you can rebuild in a failure. Just go look at http://www.raid-failure.com/raid5-failure.aspx

On 12/10/2016 at 6:59 AM, Mikensan said:

I might not be right, but I'm pretty sure Windows NIC teaming is not LACP. You could in turn just have each gigabit link assigned a different IP address, and then tell said services to listen/bind to those specific IP addresses. Now a piece of software would have a dedicated gigabit link. 

 

You have 5 gigabit ports so:

 

Port 1: plex / internet / gaming, 192.168.0.2/24 (so that you can easily share out plex to the world wide web if needed)

Port 1of4: NAS 192.168.0.3/24

Port 2of4: NVR 192.168.1.1/24

 

 

Or something like that anyway. NIC teaming / LACP will not divide up the bandwidth per se anyway, so you're still only going to get a gigabit per service/software anywho (unless you had 10+ 1080p cameras then it might be beneficial).

Windows (server at least) has multiple options, and does support LACP go have a look at https://gallery.technet.microsoft.com/windows-server-2012-r2-nic-85aa1318 for a simplified summary of nic teaming.

[Electronics Wizardy]

1 minute ago, Blake said:

No, with RAID 5 you have like a 75% chance of the array failing before you can rebuild in a failure. Just go look at http://www.raid-failure.com/raid5-failure.aspx

Those numbers are way higher than they are in real life. I have rebuilt many raid 5's with out error.

 

Also with raid 5, if you are rebuilding and get a error, it will just skip that block, not fail the rebuild. You also have backups of a array.

[leadeater]

4 minutes ago, Electronics Wizardy said:

Those numbers are way higher than they are in real life. I have rebuilt many raid 5's with out error.

 

Also with raid 5, if you are rebuilding and get a error, it will just skip that block, not fail the rebuild. You also have backups of a array.

I've rebuilt more than I can remember and built hundreds of RAID 5 arrays and not a single one has failed, not without the controller itself dying and those were systems I didn't build using 3ware controllers which I'd NEVER use.

 

No body builds RAID 5 arrays without patrol reads scheduled, like you actually have to turn it off, and that is the only time you'll ever be at a real risk of RAID 5 failure. The possibility of it is real and reasons given are accurate while also being flawed, the math is utterly wrong because it doesn't factor in patrol reads and without taking that in to account you might as well throw a dart at a dart board and where it hits is the answer to the equation.

[Electronics Wizardy]

7 minutes ago, leadeater said:

I've rebuilt more than I can remember and built hundreds of RAID 5 arrays and not a single one has failed, not without the controller itself dying and those were systems I didn't build using 3ware controllers which I'd NEVER use.

 

No body builds RAID 5 arrays without patrol reads scheduled, like you actually have to turn it off, and that is the only time you'll ever be at a real risk of RAID 5 failure. The possibility of it is real and reasons given are accurate while also being flawed, the math is utterly wrong because it doesn't factor in patrol reads and without taking that in to account you might as well throw a dart at a dart board and where it hits is the answer to the equation.

And we also backup our arrays so if something goes bad, we can restore the backups, right guys...

 

 

[Blake]

4 hours ago, Electronics Wizardy said:

Those numbers are way higher than they are in real life. I have rebuilt many raid 5's with out error.

 

Also with raid 5, if you are rebuilding and get a error, it will just skip that block, not fail the rebuild. You also have backups of a array.

Yeah, they are, but if this is just a bunch of stuff you downloaded, sure who cares if it failed. But, lets say you store you CRM/Finance database on it? yeah no.

 

Sure having a single record being corrupted or something might not, but it also might be a massive deal. then consider how the disk failed. considering most of the time disks are purchased together/at the same time from very similar batches, so.... if you loose the 2nd disk then your up shit creek without a paddle. then you need to make sure you have DR or backups. which, lets face it, most organisations (not their it departments) don't fund/invest in.

 

Also in real life you where probs using an actual enterprise disks, which is what mitigates this, they are normal 10^15+ where as WD RED 3TB disks listed are 10^14.

[leadeater]

10 minutes ago, Blake said:

Sure having a single record being corrupted or something might not, but it also might be a massive deal. then consider how the disk failed. considering most of the time disks are purchased together/at the same time from very similar batches, so.... if you loose the 2nd disk then your up shit creek without a paddle. then you need to make sure you have DR or backups. which, lets face it, most organisations (not their it departments) don't fund/invest in.

Yep I've witnessed a cascading disk failure of 3 disks (4 but 1 was able to be re-seated then replaced cleanly), what made this incident worse was it contained an entire hospital's worth of medical data so shit creek is an understatement, more like the place all the shit creeks lead in to lol.

[Electronics Wizardy]

14 minutes ago, Blake said:

But, lets say you store you CRM/Finance database on it? yeah no.

Backups and partol reads and your good.

 

Also if its that important, you should have HA, so if a server fails, your data is still accessable.

15 minutes ago, Blake said:

.. if you loose the 2nd disk then your up shit creek without a paddle. then you need to make sure you have DR or backups. which, lets face it, most organisations (not their it departments) don't fund/invest in.

If its importand to a company you have backups and HA.

 

16 minutes ago, Blake said:

erprise disks, which is what mitigates this, they are normal 10^15+ where as WD RED 3TB disks listed are 10^14.

Real life is much higher. Read errors are rare, also parity on drives lowers this chance even more.

[Mikensan]

@Blake - very informative, so by default it doesn't use LACP but something that must be set via powershell. Didn't know that was an option

 

[Blake]

15 hours ago, Electronics Wizardy said:

Backups and partol reads and your good.

 

Also if its that important, you should have HA, so if a server fails, your data is still accessable.

If its importand to a company you have backups and HA.

 

Real life is much higher. Read errors are rare, also parity on drives lowers this chance even more.

Wow...

 

1. Not everyone has partol reads. Backups are fine, but if you look at the ATO... backups still didn't protect them from having to restore 1PB of data. Your looking at a long downtime.

2. Management tend to not always make the correct decision/think that some things are acceptable risks. so may not approve HA or correct DR plans.

 

which brings me to:

Sure I do admit that these numbers are theoretical, never said you where wrong about real world, but you must be insane if your not at least going to choose more then 1 disk redundancy for anything mission critical. A lot of IT people know about HA and DR (a lot talk about it but don't actually know it), but you'll find very few of them understand that spending that $1m on a spare SQL server + SAN isn't their choice, its management. So go ahead and set up HA/DR on that imaginary Server. 

[leadeater]

22 minutes ago, Blake said:

Wow...

 

1. Not everyone has partol reads. Backups are fine, but if you look at the ATO... backups still didn't protect them from having to restore 1PB of data. Your looking at a long downtime.

2. Management tend to not always make the correct decision/think that some things are acceptable risks. so may not approve HA or correct DR plans.

 

which brings me to:

Sure I do admit that these numbers are theoretical, never said you where wrong about real world, but you must be insane if your not at least going to choose more then 1 disk redundancy for anything mission critical. A lot of IT people know about HA and DR (a lot talk about it but don't actually know it), but you'll find very few of them understand that spending that $1m on a spare SQL server + SAN isn't their choice, its management. So go ahead and set up HA/DR on that imaginary Server. 

Speaking from a hardware RAID card perspective patrol reads are on by default, I don't use mdadm for example so can't speak to those software type RAID configurations or motherboard RAID.

 

As for anything that is 1PB in size yea there is no way in hell a system like that would have 1 disk redundancy. Much older SANs have always done basic consistency checks like patrol reads but newer ones use disk pooling configured like RAID 6 but use specialty file systems and ensure consistency much like ZFS, but I'm sure you already know this.

 

Our Netapps for example have 2 parity disks + 2 hot spares per disk tray so per rack there is actually a significantly large number of disk redundancy.