Something strange is afoot at the Circle K, Ted

notarat

2[H]4U
Joined
Mar 28, 2010
Messages
2,501
So I've been backing up data from my 2 computers to each other for a while now and have encounterd odd problems...

1 - The 1600X can see and access and copy to/from the 2700X system, but the 2700X can't access the 1600X at all
2 - The copy speed when backing up one to the other is pretty slow.

I've turned off the firewall on the private network and the usernames/passwords are the same on both systems.
I've replaced network cables, tried different 1GB switches, blown out all network settings and up both from scratch.

Decided I would try to use a crossover cable (computers sit next to each other) and when I decided to do that I went ahead and grabbed a pair of Aquantia 10GB NICs

Made a CAT6A crossover cable, installed the cards and loaded the proper drivers on both systems and I'm still seeing the exact same problems. Double-checked my settings and I can't see where the problem is.

Could use some assistance in catching what I've obviously overlooked.

I have screenshots of the settings used on both computers. I know it's something simple I've overlooked so I would appreciate a second (or more) pair of eyes to spot it.

1600X settings
1600x Settings.jpg

2700X settings
2700XSettings.jpg
 
Well first things out of the way, you should leave the default gateway and DNS entries blank on the 10gb adapters.

Secondly, can you actually ping across the ten gig links? Try by name first, see what the name is resolving to, it may have either a cached entry for the 1600x or you could have modified your host file long ago and not remember.
 
Well first things out of the way, you should leave the default gateway and DNS entries blank on the 10gb adapters.

Secondly, can you actually ping across the ten gig links? Try by name first, see what the name is resolving to, it may have either a cached entry for the 1600x or you could have modified your host file long ago and not remember.

According to all info I've found online so far, leaving in or removing the gateway & DNS has no effect either way. It just has to be present on both if it's used on either (e.g. if the IP of 1600x is 192.168.0.1, then that IP needs to be in the default gateway & DNS blocks of both the 1600x and the 2700x)

I have removed them, though.

Ping times are less than 1ms when pinging each other by name (ping 1600x from 2700x or ping 2700x from 1600x)


still researching
 
And when pinging by name what IP do they resolve to?

Not sure where you read that you should have the gateway set to the IP of the PC. You shouldn't have more than one default gateway. If the 10gb adapters are connected via direct link there should be no gateway set on them.

Also, if you are dead set on using the 10gb link between them, then you need to make sure that when you access them by name that they are using that link. You have two pc's on the same network and also direct connected, if you want them to use that direct link then the name will have to resolve to an IP on that direct link. Other wise they could be pinging each other over the 1gb nic. That's why I asked what the name resolves to not just whether or not you can ping.
 
And when pinging by name what IP do they resolve to?

Not sure where you read that you should have the gateway set to the IP of the PC. You shouldn't have more than one default gateway. If the 10gb adapters are connected via direct link there should be no gateway set on them.

Also, if you are dead set on using the 10gb link between them, then you need to make sure that when you access them by name that they are using that link. You have two pc's on the same network and also direct connected, if you want them to use that direct link then the name will have to resolve to an IP on that direct link. Other wise they could be pinging each other over the 1gb nic. That's why I asked what the name resolves to not just whether or not you can ping.

As I stated before, I have removed the Gateway/DNS information. and yes, I'm dead set on using the 10GB connection between them. I routinely move 50-100GB of data between them on an almost daily basis. 1GB ain't cuttin' it.

Pinging 1600x (c:\ping 1600x) from 2700x returns the IP of 1600x(192.168.0.1)

Pinging 1600x [192.168.0.1] with 32 bytes of data:
Reply from 192.168.0.1: bytes=32 time<1ms TTL=128
Reply from 192.168.0.1: bytes=32 time<1ms TTL=128
Reply from 192.168.0.1: bytes=32 time<1ms TTL=128
Reply from 192.168.0.1: bytes=32 time<1ms TTL=128

and vice versa

I have also verified that 1600x and 2700x are using the crossover connection by unplugging my cables connected to the 1GB NICs.
 
In your screenshots you show network discovery and file sharing etc on the 'Private' profile but it doesn't show it as the 'current profile'

Are you sure those are set to the 'Private' setting and not guest or public?
 
In your screenshots you show network discovery and file sharing etc on the 'Private' profile but it doesn't show it as the 'current profile'

Are you sure those are set to the 'Private' setting and not guest or public?

While I'm not at the computer right now (I'm at work) I'm absolutely positive, 100% so, that those Aquantias are set on the "private" profile setting. I triple verified it yesterday because I thought maybe I mixed them up. I didn't. Both are indeed on the Private.

I'm totally baffled at this point and will most likely just wipe-n-reload the OS'es this weekend. Grrr....
 
According to all info I've found online so far, leaving in or removing the gateway & DNS has no effect either way. It just has to be present on both if it's used on either (e.g. if the IP of 1600x is 192.168.0.1, then that IP needs to be in the default gateway & DNS blocks of both the 1600x and the 2700x)

I have removed them, though.

Ping times are less than 1ms when pinging each other by name (ping 1600x from 2700x or ping 2700x from 1600x)


still researching

That is incorrect. If you have 2 adapters, you only set the default gateway on the adapter used for non-local (internet) traffic. Especially if the 10G lan is directly connected between the 2 machines. You do need to configure the second 10G adapter with a completely different CIDR range. If you are using 192.168.0.0/24 on your main adapter, use 192.168.10.0/24 on the 10G and give each machine a static IP in that range. Any traffic with a destination of the 10G CIDR will use that adapter, all other traffic will use your main adapter.

If it works one way, but not the other, it sounds like something is blocking it. Make sure your adapter is using a real profile, preferably 'private' or you can have issues even with the FW turned off. You can turn on network discover and filesharing which should automatically enable the correct rules on the FW.

Can you hit the share on the box itself? ie if you can't reach serverB from serverA, can serverA reach \\<ServerA_IP>


*edit* Forgot to add, crossover cables are no longer needed with 1G+ network hardware. They added that to the standard, so the network adapter itself will do the crossover if required.
 
If you look at OP's screen shots he does have the adapters on different subnets.

I agree though, if it's working one way now and not the other, and you can ping across both ways, this is looking like some kind of windows setting.

Check and make sure all the ssdp crap is running, I forget all the services you need for shares to be discoverable on the network through windows automagically but there's three or four
 
Didn't see the images when I first loaded the post, but it looks like you are using host names. Do you have an internal DNS server, or static host overrides setup on your firewall for them? If not you're asking for issues and should be using the IP address. That will remove the need for the windows network discovery (still need file sharing rules enabled though). You can map the shares on each machine and have it auto connect at boot to make it easier.

On the box with shares you can access, run Resource Monitor. Watch the network tab and check the TCP connections list. Then attempt to open the share from the other machine. Do you see the IP listed in the TCP list? That will confirm the connection is making it to the box on the correct port. Does it stay open, or close immediately and go grey?
 
Didn't see the images when I first loaded the post, but it looks like you are using host names. Do you have an internal DNS server, or static host overrides setup on your firewall for them? If not you're asking for issues and should be using the IP address. That will remove the need for the windows network discovery (still need file sharing rules enabled though). You can map the shares on each machine and have it auto connect at boot to make it easier.

On the box with shares you can access, run Resource Monitor. Watch the network tab and check the TCP connections list. Then attempt to open the share from the other machine. Do you see the IP listed in the TCP list? That will confirm the connection is making it to the box on the correct port. Does it stay open, or close immediately and go grey?
I even seen a situation with a local DNS provider that experienced weird issues with speed, connecting and disconnecting, and other random stuff... A misconfigured DNS can be worse than not having one. If memory serves me correct, this setting was something about allowing DHCP clients to register their hostname. I didn't dig too deep, but it appeared that some devices simply aren't expecting to register their hostname when given a DHCP address, and their was some limbo period during this negotiation where the devices wouldn't communicate with each other except for DHCP/DNS ACK and REQ. It seemed like the devices would get confused. Eventually, they would get past that limbo point and have network and internet access. Sporadically, this limbo would happen again and either cause the client device to get kicked and/or have slow speeds.

To OP, this may or may not pertain to your situation, but I'm posting hoping that it is useful to you. Because, DNS can be fuckey.
Also, if you aren't using IPv6, I recommend turning it off completely. Have definitely seen that cause issues too.
 
That is incorrect. If you have 2 adapters, you only set the default gateway on the adapter used for non-local (internet) traffic. Especially if the 10G lan is directly connected between the 2 machines. You do need to configure the second 10G adapter with a completely different CIDR range. If you are using 192.168.0.0/24 on your main adapter, use 192.168.10.0/24 on the 10G and give each machine a static IP in that range. Any traffic with a destination of the 10G CIDR will use that adapter, all other traffic will use your main adapter.

If it works one way, but not the other, it sounds like something is blocking it. Make sure your adapter is using a real profile, preferably 'private' or you can have issues even with the FW turned off. You can turn on network discover and filesharing which should automatically enable the correct rules on the FW.

Can you hit the share on the box itself? ie if you can't reach serverB from serverA, can serverA reach \\<ServerA_IP>


*edit* Forgot to add, crossover cables are no longer needed with 1G+ network hardware. They added that to the standard, so the network adapter itself will do the crossover if required.

If you look at OP's screen shots he does have the adapters on different subnets.

I agree though, if it's working one way now and not the other, and you can ping across both ways, this is looking like some kind of windows setting.

Check and make sure all the ssdp crap is running, I forget all the services you need for shares to be discoverable on the network through windows automagically but there's three or four

I even seen a situation with a local DNS provider that experienced weird issues with speed, connecting and disconnecting, and other random stuff... A misconfigured DNS can be worse than not having one. If memory serves me correct, this setting was something about allowing DHCP clients to register their hostname. I didn't dig too deep, but it appeared that some devices simply aren't expecting to register their hostname when given a DHCP address, and their was some limbo period during this negotiation where the devices wouldn't communicate with each other except for DHCP/DNS ACK and REQ. It seemed like the devices would get confused. Eventually, they would get past that limbo point and have network and internet access. Sporadically, this limbo would happen again and either cause the client device to get kicked and/or have slow speeds.

To OP, this may or may not pertain to your situation, but I'm posting hoping that it is useful to you. Because, DNS can be fuckey.
Also, if you aren't using IPv6, I recommend turning it off completely. Have definitely seen that cause issues too.

Updated:

1 - Thanks for the assistance. I've spent 2 days following your suggestions, but it didn't help (not your fault...see below)
2 - It seems that, for some strange reason there was aome arcane registry setting or something (I certainly don't remember tweaking anything related to networking in the registry) that was causing the issue.
3 - I finally reached a point Thursday night where I decided it would be better to slick and reload than fight it any longer.
4 - I'm thinking over the last year or so (in the version upgrade processes each went through at separate times) something got borked. I reloaded both computers from scratch with the same 1803 installation media. Loaded the same NIC Drivers, same video drivers, same sound drivers, etc and each computer was now unable to even see each other. (yeah...it got worse before it got better)

The errors I got after step 4 were ones I'd seen before, though, and I had a good idea how to correct them. After futzing around for a while, low and behold, they can see and access each other fine, on the 10GB cards.

Copy speeds on my 13Gb file transfers aren't "horrible" from Mechanical HDD to Mechanical HDD, but they aren't what I was hoping for, lol

2700X_to_1600x_thruput.jpg


However, copying 13GB from the SSD on my 2700x to the EVO M.2 on 1600x seems to be the speed I was looking for in the first place.

2700X_to_1600x_SSD_thruput.jpg


So...now all I need are about TEN 4TB SSDs, lol
 
Wow, no one suggested GPO or registry changes? That would have fixed it right up after a reboot.

For future reference:
gpedit.msc
Computer > Admin Templ > Network > Link-Layer Top Disc
- Enable LLTDIO (Enable in Domain and Public if your current network connection is ID'd as such)
- Enable RSDPNDR (Enable in Domain and Public if your current network connection is ID'd as such)

If you have Windows Firewall running (which you probably do)
gpedit.msc
Computer > Policies > Windows Settings > Security Settings > Win Firewall > Win Firewall w/Adv > Inbound Rule
- Create new rule > Predefined > Network Discovery > (select all) > Allow

Either run a gpupdate /force or reboot your computer.

To check type of connection:
Computer\HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\NetworkList\Profiles\
Each connection will be in {23409230856543t} format. Click through each one to see which one applies.
The registry key "Category" is what you are looking for.
0 = Public
1 = Private
 
Last edited:
Wow, no one suggested GPO or registry changes? That would have fixed it right up after a reboot.

For future reference:
gpedit.msc
Computer > Admin Templ > Network > Link-Layer Top Disc
- Enable LLTDIO (Enable in Domain and Public if your current network connection is ID'd as such)
- Enable RSDPNDR (Enable in Domain and Public if your current network connection is ID'd as such)

If you have Windows Firewall running (which you probably do)
gpedit.msc
Computer > Policies > Windows Settings > Security Settings > Win Firewall > Win Firewall w/Adv > Inbound Rule
- Create new rule > Predefined > Network Discovery > (select all) > Allow

Either run a gpupdate /force or reboot your computer.

To check type of connection:
Computer\HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\NetworkList\Profiles\
Each connection will be in {23409230856543t} format. Click through each one to see which one applies.
The registry key "Category" is what you are looking for.
0 = Public
1 = Private
Ain't nobody got time fo dat
 
They're (4TB SSDs) not as expensive as I'd thought. $8k for 10 Samsung 860EVOs...hmm...

Still 1200% or so per gigabyte over spinning storage though. Fine for smaller arrays that need the response times and/or IOPS, but really not much utility for a homelabber like myself ;).

[and as relatively expensive as the storage is, that's attainable- I shudder at the pricing of infrastructure that could really make it useful; has 100Gbit hit eBay in force yet?]
 
Back
Top