The Hivemind is a cluster of computers used for experimental purposes, that is either parallel networking, networking, and things alike. So far it is at its early stages as storage, CPU and memory upgrades are planned on those machines.
The Hivemind currently have 4 components picked up from garbage or second-hand. Therefore they may be underpowered but should be enough for now. It is possible on some of those machines to upgrade up to a 3rd-gen Intel Core CPU.
They are connected together in their own LAN, the master is not supposed to compute and is only here to supervise and control the slave machines.
No upgrade are really planned for now. I am mostly using off-the-shelf parts and things alike. But if this machine is frequently used, RAM, CPU and storage will be upgraded.
One of the nodes might also be used as a secondary data hoarding file server, to make use of old parts lying around.
The Dell Precision T1600 towers can be upgraded with up to 32 GB of DDR3 RAM and 3rd gen Core i7s (giving us 8 threads to work with). It may also support old GeFore GPUs, so GPU upgrade may be possible as well.
The Dell Optiplex 960 is older though and supports only old Pentium 4 and Core 2 CPUs. So an upgrade to a Core 2 Quad may be desirable.
The only potential upgrade for the master is to buy an ARM computer with dual NIC as a router and command center, no computation work is planned on this machine to ensure network capabilities of the cluster. It will be used through SSH rather than using a laptop.
So far, the Hivemind 03 is being upgraded with a Xeon E3-1280 (8 threads at 3.90GHz max!), which is a server/workstation CPU supported by the socket in the machine, 16 gigs of DDR3 RAM and a GeForce GTX 750 for GPU calculation.
the Hivemind 02 is going to have a similar upgrade. I am looking forward to see if I can pick up some GPU or have to buy one. Buying a Quadro 2000 could be a desirable thing to do.
The Hivemind 01 may receive RAM upgrade and eventually a CPU upgrade, but for a more power-efficient CPU rather than a more powerful one. Considering that the CPU socket (LGA 775) is even more an antiquity than the LGA 1155 on the other nodes, efforts will be made to turn it into a NAS instead. Maybe mobo replacement will be considered if it is worth being a NAS.
After upgrade, the specs should be the following:
As it is a Beowulf-like cluster for parallel computing, it has a master-slave architecture to work with stuff. However, while a lot of those clusters still make use of the master's computing power to do the computation stuff, mine does not. The Hivemind Master is not very powerful (it's my high school laptop with a Celeron in it) and is simply used as a control center.
The fact that the master does not do any tough work however makes the master-slave analogy more accurate. See: capitalism
This is the network I decided to set up for The Hivemind project. It has its own LAN rather than use the LAN of my home router. It allows more flexibility as it is also an experimental networking setup. It allows broadcast packets to be sent only between the nodes of the cluster rather than flooding the broadcast traffic of my home network. This also allows more flexibility in local domain name and IP address allocation as it is limited to its own LAN. I can also flood the network without fearing it obliterates my internet connection out of existence.
( 192.168.2.0/24 ) ( some local subnet ) ( 0.0.0.0/0 )
[ HM01 ] \ NAT
[ HM02 ] |=-= [ MASTER ] =-= [ (my local network) ] =-= [ internet ]
[ .... ] /
All machines have Debian Linux installed, but PXE is also enabled in case a change of decision is made. Setting up those machines one by one was a tedious and annoying process.
Computing will be done using the Message Passing Interface (MPI), specifically
mpich.
The Hivemind Master does not do computer work. However, it works as a control center and also a router. It offers DHCP, NAT, DNS and NFS services for the slave nodes so they have a local domain name, an IP address defined by their MAC address, a common file system and access to internet (which can be taken down if necessary). It may also offer PXE network booting if necessary later on.
The Master also handles Wake-on-LAN to allow remotely turning on and off the machines (it's very convenient). More tooling will be created over time as I use it.
They're simple Debian Linux machines controlled by the master. They are accessible through SSH by the master. There's not much stuff to say about them other than they're (supposedly) powerful and compute stuff.
But since they're trash-picked there had some stuff to do such as removing RAM modules that does not work, replacing the CMOS battery and removing those god damn BIOS passwords.
Here's some computations done on the machine using mpi:
The program is in the mpi page. It is the first program used on The Hivemind.
Number of iterations: 10e11
Results:
pi = 3.1415926535311076150946973939426243305206 Time taken: 315
The results are not super impressive and my laptop is probably faster at doing it. But I'm still happy of the results. As stated earlier, upgrades are possible.
There is not really a specific purpose motivating this experimental setup other than playing with networking and parallel computing.