cluster switch

10gb gpunet and aeronomy processing

Connections and locations of switches

• SSR8 - Main ao switch
• SSR8 is the main ao switch.
  
• It has room for 14 plugin modules. Modules can be 100 Mbit or 1 Gigabit holding 1,2,4, or 8 inputs.
  
• Notes: P is the switch port number (labeled on the front of the switch).
• N3 10Gbit Sw
• Main 10Gbit switch. Has room for 2*3 10 gbit connections.
• Currently have modules for 2*2 10Gbit connections.
• Located in computer room by exit to reciever room.
• C3kserver 10Gbit Sw:
• 2*10 Gbit input
  
• 24 1Gbit inputs
• Located in computer room  between clockroom and reciever room doors.
• compute servers GalfaSn, etc connected to it.
• C3koffice 10Gbit Sw:
• 2*10 Gbit input
  
• 24 1Gbit inputs
• Located building 2 3rd floor.
• Eventually computers in offices will be connected to it.
• C3kmock 10Gbit Sw:
• 2*10 Gbit input
  
• 24 1gbit inputs
• Located in clock room.
  
• Mock spectrometer will eventually be connected to it.
  

• Port3:
• 3.1 trabucco
• 3.2 3com switch computer room
• 3.3 galfa
• 3.4 JB17-backend-east
• 3.5 lab
• 3.6 redSeismica
• 3.7 Hilltop Receiver.
• 3.8 No Xmit
  
• Port5:
• 5.1 PRTCFT1
• 5.2 Learning center
• 5.3 ControlRoom North
• 5.4 Modem server
• 5.5 Remote2+DNSext
• 5.6 Brandon
• 5.7 JB11-Backend-East
• 5.8 CORREO ???also on 8.1
  
• Port8:
• 8.1 almacen
• 8.2 Visitor center
• 8.3 optical
• 8.4 compras
• 8.5 lidar
• 8.6 wappswitch
• 8.7 candystore-else5
• 8.8 vsq 1-6

Cluster switch PC6024

    The cluster switch has a number of important servers plugged into it. It is plugged into the main switch via:

• cluster switch pc6024 (rack R5) -> pc5224 switch (Rack R3) -> gi.9.2 of main switch

Using the gpus and the 10gb network

    A number of servers have been connected on the gpunet 10gb network. They are used in the processing of

the aeronomy clp data.  The table below shows the servers, their ip addresses, and the interface names
common colors are on the same network connections... (white/aonet probably goes through more switches).
Some notes:

•  There is (yet) no inter-network routing. You need to make sure that both cpu's have the same colored network.
  
• The network throughput is probably determined by the spd column (1 or 10gb).
• Note that the 10.10.11.9 on tstor(yellow) is 10gb, while all of the other yellow interfaces are 1gb.
• the switch from 10 to 1gb happens at the switch. So if multiple yellow cpus are pulling data from 10.10.11.9, then you can get more than 1gb/sec
• I've noticed that on say on gpuserv 1..
  
• cd /net/10.10.11.5/export/GPUSERV3/usrpdata
• then
  
• cd /net/10.10.10.3/export/GPUSERV3/usrpdata
• df  -kh shows:
• Filesystem                                Size  Used Avail Use% Mounted on
  10.10.11.5:/export/GPUSERV3/usrpdata      40T   39T  1.4T  97% /net/10.10.11.5/export/GPUSERV3/usrpdata
  10.10.11.5:/export/GPUSERV3/usrpdata      40T   39T  1.4T  97% /net/10.10.10.3/export/GPUSERV3/usrpdata
• They both come from the first interface selected.

last update: 14apr19

home_~phil