Results 1 to 5 of 5
  1. #1
    Lounger
    Join Date
    Dec 2009
    Location
    Wellington NZ
    Posts
    32
    Thanks
    6
    Thanked 0 Times in 0 Posts

    Subnet transfer speed issues

    Because I work from home I have divided my home network into two subnets, for simplicity's sake I'll call them Work and Home. I have turned off DHCP on the modem/router supplied by my ISP and added a raspberry pi as my DHCP server. I've attached a schematic of the network which shows how several devices are connected together. (The actual network is quite a bit more complex than this but I've reproduced the problems I'm having with the cut-down version shown in this diagram.)


    All cables are cat6, both hubs are Dlink DGS1008D (hence gigabit), and every device is connected via a gigabit network connection except for the pi which is 100M.


    I am having three issues which are probably related but wonder if anyone can help me with an explanation for what I'm seeing so that I can address them.


    1. File transfer speeds are very different depending on the subnet.
    a. Transfers between 192.168.0.4 (a Synology Diskstation) and 192.168.0.21 (my homebuilt desktop machine) run at 100Mb/s.
    b. Transfers between 172.16.9.5 and 172.16.9.4 run at 10Mb/s.
    c. Transfers between 172.16.9.5 and 192.168.0.4 run at 1Mb/s.


    2. Although the pi is not on the same hub as either the XPS15 or the Diskstation, when copying files between them - (c) above - the pi receives approximately 10,000 IRQ32 interrupts per second. For comparison, when copying within either subnet - (a) or (b) above - the pi receives approximately 700 interrupts per second, which is the same number as it receives when the network is idle.


    3. I don't know whether it is relevant or not but the only difference between the two subnets is that the DHCP service on the pi points machines on the Home subnet to do DNS lookups externally (using my ISP's name server or Google's as a fallback), while machines on the Work subnet use the pi so that they can resolve addresses in the VPN which my work machines are connected to (the pi is running openVPN to enable this). But even if the only activity in the network is copying my backup files from the work XPS15 to the home Diskstation, the 'named' (DNS) and 'openvpn' processes on the pi are CPU bound. When the backup finishes, they revert to 1% cpu at most. This means that when the backup *is* running any other network activity in the work subnet is prone to timeout, and connections to other sites in the VPN are dropped.


    Issues 2 and 3 would cease to be problems if the backup transferred at the speed it should be capable of.


    Anyone got any helpful ideas?

    Cheers
    T
    Attached Images Attached Images

  2. #2
    Lounger
    Join Date
    Dec 2009
    Location
    Wellington NZ
    Posts
    32
    Thanks
    6
    Thanked 0 Times in 0 Posts
    A few additional observations.

    1. I called them hubs but they are ethernet "switches", sorry for any confusion.

    2. "Work" machines are allocated the following by the DHCP service:
    Default gateway: 172.16.9.1
    DHCP server: 192.168.0.1
    DNS servers: 172.16.9.1

    "Home" machines get this:
    Default gateway: 192.168.0.254
    DHCP server: 192.168.0.1
    DNS servers: 8.8.8.8, 8.8.1.1

    3. Traceroute confirms that home machines can't see work ones, and that work machines can only see home machines via the pi.
    From 192.168.0.21
    tracert 172.16.9.5
    1 2 ms 3 ms 3 ms 192.168.0.254
    2 ~ ~ ~ Request timed out.
    3 ~ ~ ~ Request timed out.
    tracert 192.168.0.4
    1 <1 ms <1 ms <1 ms 192.168.0.4

    From 172.16.9.5
    tracert 172.16.9.4
    1 <1 ms <1 ms <1 ms 172.16.9.4
    tracert 192.168.0.4
    1 <1 ms <1 ms <1 ms 172.16.9.1
    2 1 ms 1 ms <1 ms diskstation.home [192.168.0.4]

    That's the cause of my performance issue, because despite the gigabit capability everywhere else cross-subnet transfers are all channeled through the pi bottleneck. But how can I avoid that?

  3. #3
    WS Lounge VIP Browni's Avatar
    Join Date
    Dec 2009
    Location
    Rochdale, UK
    Posts
    1,650
    Thanks
    38
    Thanked 161 Times in 139 Posts
    Is 8.8.1.1 a typo?

  4. #4
    Lounger
    Join Date
    Dec 2009
    Location
    Wellington NZ
    Posts
    32
    Thanks
    6
    Thanked 0 Times in 0 Posts
    Quote Originally Posted by Browni View Post
    Is 8.8.1.1 a typo?
    I think it's a hangover. Google's public DNS used to be 8.8.8.8 and 8.8.1.1. (It seems that might not be so any more? 8.8.4.4 instead?) My ISP's DNS was always dropping offline so I switched to Google several years ago.

  5. #5
    WS Lounge VIP
    Join Date
    Dec 2009
    Location
    Earth
    Posts
    8,176
    Thanks
    47
    Thanked 982 Times in 912 Posts
    It's a routing issue - the pi is doing your routing instead of the switches. Basically, you can't do routing via "normal" switches so your network is a not-work.

    Hub 2 needs to be a router / layer 3 switch and only work machines should be connected to it. (Even doing this you have work data traversing the home network).
    Hub 2 would have an IP in the 172 range and work machines would then have a default gateway of the router.
    Hub 2 would need a "helper" address to forward DHCP requests to the pi.

    cheers, Paul

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •