Project Home Lab: Network Decisions
So far in the series, I’ve talked about the goals and what hardware I want to use. In this post, I’m going to talk about how I plan to connect it all together and how I’m going to get it talking to the outside world via my existing production home network.
This series will consist of the following posts. I will update the table of contents links in each post as I produce and publish the articles.
- Project Home Lab: Goals
- Project Home Lab: Existing Infrastructure
- Project Home Lab: Hardware Decisions
- Project Home Lab: Network Decisions
- Project Home Lab: Shopping List
Hyper-V to Storage
I’ve got two new servers we know that much as planned so far. The data will be on one server, the processing power on another so I need a way to interconnect them. I also need to be conscious of ensuring that whatever I deploy for the interconnect can scale up with other areas if I elect to add another host later. Most importantly, I need to be sensitive to my existing network. Me hammering away with data transfers in the lab should not under any circumstances impact my home network as getting in trouble with the wife stopping her from being able to stream videos in Xbox Fitness or play the latest Facebook craze just isn’t worth it.
We know already that I am going to need three Ethernet ports, two gigabit and one 100Mbps for the two servers to operate the on-board network ports which I have said previously I will use for management and IPMI access. This leaves the most important aspect of getting data between the two. 100Mbps is gone, never to be seen again for anything other than out of band type connections so my options are 1GbE, 10GbE or Infiniband with RDMA.
As we know, this is a home project. Infiniband requires specialist knowledge, some of which I possess from work with Xsigo in a former role and whilst yes, 40Gb/s or more between the machines would be nice, the Infiniband host bus adapters (HBAs) are expensive and the Infiniband switches even more so. 10GbE is more common however as it is still pretty much at the pinnacle of Ethernet based networking with enterprises only really taking it by the horns today it too is also very expensive which leaves me with 1GbE.
Gigabit Ethernet has been around the block a few times, parts are common and reasonably affordable. Gigabit Ethernet can be run over standard Cat 5e or as I have, Cat 6 cable so I’m reusing my existing investment in cabling and tooling for producing cables. Gigabit Ethernet also means I’m working with a single connectivity medium throughout making the identification of faults and troubleshooting simpler.
I want to get good performance out this lab so after some discussion with @LupoLoopy, we came to the decision that I should use SMB Multi-Channel, the new feature in Windows Server 2012 R2. With four ports of Gigabit Ethernet I will get decent performance at a low price and it’s easy enough to add another card to the server to open up more ports if I need later. A quad port Intel PCI Express adapter comes in at between £50 and £100 on eBay used. I got both the cards for the Hyper-V server and the storage server for £50 so make sure to keep your eye on the available items for a bargain.
I will run my Hyper-V virtual networking over these ports also and using Storage QoS in Hyper-V I can ensure that I get the right amount of storage throughput at all times.
With it now decided I’m going to use four ports of Gigabit Ethernet for my SMB Multi-Channel storage traffic and three ports for management and IPMI, I need to provision seven Ethernet ports per server. With two servers right now, that’s 14 ports and if I allow an additional seven ports for a possible future expansion, that’s 21 ports, nearly a 24 port switch full.
My current core switch, a 24 port TP-Link TL-SG3424 has about 12 ports free right now so not enough for this project. Going back to my previous statements, I want to keep any of this traffic from harming my home network performance, therefore put two and two together and you can see I’ll need a new switch for this. I don’t want to have to replace my core switch as it works perfectly well, performs well, silent and so forth. As I want to completely isolate this lab, I’m going instead to add a second switch to my network for the lab and I will trunk the lab up to the core for internet access. With this leaf switch design for the network, the only traffic that needs to leave or enter the core switch to and from the lab is external access from myself or Internet access requests, containing the storage traffic and protecting my home interests.
I looked at all the options and came to the swift conclusion that I was going to be best placed to get another TP-Link TL-SG3424, the same as I have already for the leaf switch. 24 Gigabit Ethernet ports suit all my needs, I know it performs well, leaves me with enough ports free for an additional host in the future plus a few ports for uplinks into the core.
I wrote a review of the TP-Link TL-SG3210 I use as my access switch which has equal features and interfaces to the TL-SG3424 just it has 8 instead of 24 ports.
Access into the lab will primarily be over Remote Desktop Protocol from the home network. To do this, I’m going to be accessing the lab across uplink ports that I will configure between the core and the lab switch. The lab will be in a separate VLAN to protect the home network from any broadcasts or such like going on in the lab. As my TP-Link switches are Layer 2, the Cisco ASA will be acting as my Layer 3 router between the home network and the lab which will allow me to place IP restrictions on who can traverse from the home network into the lab.
The cost for the new TP-Link switch is about £120. I’ve already got all the tools and cable I need to wire up the networking so there is no new costs there making this arguably, the cheapest part of the project. Time is actually going to be the biggest cost factor with the networking because of the time it’s going to take me to configure all of the new VLANs for the management, VM traffic and SMB Multi-Channel traffic, the sour side of using TP-Link over Cisco and not being able to use VLAN Trunking Protocol (VTP), a feature on Cisco which I love dearly.
Thankfully, VLAN configuration is a one time thing though, so although I’ll lose a couple of hours to all the network configuration initially, the cost of buying the switches and the low power consumption of the passive cooled TP-Link devices is worth it long term.
Next up, I will do a summary post in the form of a shopping list to get down everything I’m going to be using for the project and then I’ll be heading into build.