Server Build (GitLab, Build server, VM archive / operation)

[TCWilliamson]

Hello Everyone,

 

I'm working for a company that is moving into development of their own hardware based products.

We will be dealing with PCB design & compiling for embedded linux, We have some software we wish to run and host. I would like some recommendations on a server setup.

 

Software we plan to operate with are:

Gitlab

Confluence Documentation system

NFS / FTP

VM Setup / Archiving

custom build scripts

CAD Model storage

License servers

 

 

Day-to-day operation would be storing source code from employees in Gitlab

software builds will need to be archived with Virtual machines backed up for each major revision (to counteract any software issues)

Confluence will be used to document our products and for internal task keeping.

NFS/FTP access for our developer's archive.

We will be operating buildscripts for Linux, RTOS and FPGA software.

 

I'm hoping someone would be able to offer me suggestions and support as to what questions I should be asking myself in order to work out just how much computing resources / storage / RAID / Backup we need...

 

We are currently 2 employees working on the development, however will be scaling up in the next few years as things kick off.

 

Thanks in advance for your help.

 

-Tom

[Billy_Mays]

Budget? Country?

 

Build and its in the US PCpartpicker 

       https://pcpartpicker.com/list/qZ3cKZ

[TheComputerdude]

18 minutes ago, Billy_Mays said:

Budget? Country?

 

Build and its in the US PCpartpicker 

       https://pcpartpicker.com/list/qZ3cKZ

This build has multiple issues for a server setup 

1. Redundancy one single archive drive is a bad idea if something goes wrong you are screwed

2. archive drives are so so so so so sooooo slow 

3. The fact that this is not really a server so much as a workstation build.

What you need is a proper server with an SSD boot in raid 1 or 10 and an  HDD  raid 6 with 6 10 tb drives which will provide good speed and redundancy with 40tb of storage and plenty of room for upgradeability if you get a good server chassis with enough drive bays. This may lead to you needing a raid card  .  As far as the cpu goes it is probably ok though I would recommend upgrading to a 10 core variant just to be safe. Also use a redundant power supply + ups  If you are super worried about data security it might be worthwhile to set up another less powerful server to be used as a backup server. This is a good time to use those archive drives in raid 1. If you can put this backup server somewhere offsite. But the backupserver is not super necessary 

 

Hope this helps 

[Billy_Mays]

Just now, TheComputerdude said:

This build has multiple issues for a server setup 

1. Redundancy one single archive drive is a bad idea if something goes wrong you are screwed

2. archive drives are so so so so so sooooo slow 

3. The fact that this is not really a server so much as a workstation build.

What you need is a proper server with an SSD boot in raid 1 or 10 and an  HDD  raid 6 with 6 10 tb drives which will provide good speed and redundancy with 40tb of storage and plenty of room for upgradeability if you get a good server chassis with enough drive bays. This may lead to you needing a raid card  .  As far as the cpu goes it is probably ok though I would recommend upgrading to a 10 core variant just to be safe. Also use a redundant power supply + ups  If you are super worried about data security it might be worthwhile to set up another less powerful server to be used as a backup server. This is a good time to use those archive drives in raid 1. If you can put this backup server somewhere offsite. But the backupserver is not super necessary 

 

Hope this helps 

I'm not that well for building servers and he didnt specify a budget

[TheComputerdude]

1 minute ago, Billy_Mays said:

I'm not that well for building servers and he didnt specify a budget

yeah budget will make or break this build tbh

[Elerek]

If budget is tight, ebay always has tons of older 1 and 2u dell servers for pretty cheap.

[Elerek]

Aside from "vm setup" and "custom build scripts" all it looks like you need is a good storage server with enough cpu power for git calculations.

 

What exactly is "vm setup" and how intensive are these build scripts?

[Elerek]

This + some harddrives is probably all you need. Throw some more ram in if you want, used ddr3 ecc ram is dirt cheap on ebay.

 

http://www.ebay.com/itm/Dell-R710-2U-LFF-3-5-2x-Xeon-E5520-2-26GHz-QC-48GB-No-HDD-PERC6i-Rails-/131819112582?hash=item1eb1080886:g:PCkAAOSwXeJXdD4E

 

edit:

 

The server I linked above has 18 ram slots, but only 12 sticks of 4gb ram. You could top it off with another 6x4gb for $45:

http://www.ebay.com/itm/24GB-6X4GB-DDR3-ECC-REG-MEMORY-FOR-DELL-PRECISION-WORKSTATION-T5500-T7500-/132110513471?hash=item1ec266753f:g:LlgAAOSwCU1YtOqI

[MysticalGnome]

So do you want to run everything virtual or some components on bare metal?

Are you also going to do issue tracking with Jira?

You could go the hypervisor route and deploy a pool of servers with XenServer or ESXi.

This will also enable you to migrate VM's from one host to another without having to reboot your VM's as long as the operating system of the VM is supported.

An added bonus is that you will better utilize your server capacity as running confluence (ours has 6 cores and 12GB of RAM) or Gitlab (4 cores 8GB of RAM) on a single server would be a total wast of hardware idle time.

A shared storage (NAS) or SAN (NFS or iSCSI) would be recommended but is not a requirement for this to work.
 

In terms of setups you can go in any direction really, but your budget will mostly decide what you can do.
So if you want a hypervisor setup I would go for something like this.

• Buy 3-4 older Dell PE R610/R710 servers (around 450-750 per server depending on the configuration)
  • Dual Hexa or Octo core CPU
  • Go with quad 1Gb or dual 10Gb NICs per server
  • 48GB or more RAM per server
  • 4 600GB/900GB SAS 10/15K disks (make a raid10 volume) (Smaller disks if you go for a NAS or SAN)
• Buy at least two stackable switches but only those that can be expanded in the future.
  • HPE 5130-48G 4SFP+ EI (stacks up to 9 switches with IRF and has 4x 10Gbit stacking ports) they cost around 1.5K per switch
  • HP FlexFabric 5700-32XGT-8XG-2QSFP+ (40 port 10GbE switch, same features as de 5130 but has QSFP+ for stacking) they cost around 2.5/3.5K per switch
• Buy a SAN for running your VM's off (you can make snapshots, most SAN's can do this)
  • HPE MSA 2040 which is expandable with additional units and is not ridiculously expensive and supports active dual controllers (8-16K depending on the config)
• Buy a NAS devices for backups and archival and you can run VM's on it as well with NFS or iSCSI
  • Netgear ReadyNAS (12-24TB) some models have redundant PSU (3-8K) and support cloning to another ReadyNAS device.

This setup is fully expandable and can last you a long while, but since you are starting small 3 servers, two 48 port 1GbE switches and a NAS would get you going for now.
So in total you'd have to invest about 6 to 10K to get you going.

[TCWilliamson]

Hello Everyone,

 

Thanks for the responses. Pardon the delay responding. I've been very busy doing site visits.

 

You've given me a lot to think about, admittedly some of those systems seem a lot larger then what will be needed for a small development team. I will have a chat with our IT service provider as well when they come in next.

 

I've been also advised that the compiling process for the FPGA hardware we use often has a lot of single-core processing so a high-speed (overclocked) single core holds more benefit in aspects then many cores.

 

 

Regards,

Thomas Williamson

[Mattias Edeslatt]

Server and overclock should never exist in the same sentence  

 

IX Systems have a good reputation, at least in the FreeBSD/Linux-community. They customize to your needs. Otherwise I would call Dell and talk to their server reps. I wouldn't build my own in a production environment and to handle all troubleshooting, support and configuration. What are your cost if you have downtime? Dell offers same day/24-hours replacement and on-site support.

 

Don't forget that you need backup and maybe even redundancy. UPS is a must.

Hope this helps.

[Mikensan]

You could honestly host a majority of that off a single VM with the rest underneath a docker to really save money.

 

Budget plays a huge role, if you go with vmware their essentials package is around $600? Not to mention if you use Microsoft for your server o/s is also $600/server(16cores). Licensing is a large cost, something to think about.

 

Other is who is going to be dedicated to keeping this ship afloat? If you do a custom build it's going to require tinkering throughout its lifespan, so a turnkey product may be a better solution.

 

A popular move these days is put your storage + hypervisor into one system. You could setup something like Proxmox + ZFS and have a really solid long lasting setup.