Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I use Proxmox on a server where it manages a herd of containers plus a few VMs for those times I need to run Windows. On a single DL380G7 Proxmox shepherds the following:

Containers:

- router (OpenWRT)

- mail

- web

- media (serves audio, video, books, photos etc)

- database

- authentication, central letsencrypt instance

- remote desktop/app session

- build services

- hydra (NSA's decompiler)

- peer to peer apps

- 'cloud' (Nextcloud)

- comms (Jitsi meet, Nextcloud Talk + related services)

VM:

- Windows, several versions for experimenting

- ELSA (VAG online service manual)

All on a single box, easily managed through either the Proxmox web interface or - for those things which the web interface does not facilitate - through the CLI tools or by editing LXC or QEMU config files.

While there are some areas where I have had to intervene to make things behave the way I want them to behave (e.g. I don't want to use ZFS since I prefer LVM raid so I had to set up storage 'by hand', the snapshot backup system does not cope with FUSE filesystems so I have to use hook scripts to unmount those before backup and remount them afterwards, etc) all in all the experience with Proxmox has been mostly positive.



Impressive. Can you give a rough guideline about how much men and CPI you allocate to each task (and how much physical men and CPU are in the host)?


I assume that by 'men' you mean 'cores'? If so, the host has two 6-core/12-thread CPU's (X5675@3.07GHz) and 128GB of memory, the internal array is populated with 8x146GB, the machine is hooked up to a NetApp DS4243 with 24x600GB (20 of which are in use, 4 cold spares). The machine has plenty of spare capacity, both CPU-wise as well as memory-wise. Storage is getting a bit short, eventually I'll start swapping drives in the DS4243 with higher-capacity versions. The whole contraption is housed in a sound-proofed cabinet with equipment in top, produce drying racks in the bottom. A forced draft fan all the way in the bottom of the rack draws in air through a large truck air filter in the top. This was the spare heat is used for a productive purpose instead of just wasted.

Resources are used as needed, when a container seems to be running close to capacity I add cores, memory or storage. In the current configuration I assigned 2 cores/1GB to mail, 2 cores/1GB to auth, 8 cores/8GB to build, 6 cores/16GB to (data)base, 8 cores/16GB to serve (which runs a host of web services), 6 cores/8GB to session (which runs remote desktop/single app sessions through X2go), 4 cores/1GB to panopticon (CCTV), 2 cores/1GB to p2p, 4 cores/4GB to comms and 4 cores/512MB to router. Total load on the machine is negligible, memory pressure is... absent (currently ~8GB in use, ~84GB buffer/cache, ~36GB free).

A machine like this one can be had for not that much ex-lease. As long as you get a G7 or newer power usage is acceptable, G6 and under are power hogs.


Thanks. I was writing on mobile, and autocorrect corrected “mem” as in memory to men without me noticing.... sorry.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: