Huawei Cloud Cashback Credits Tencent Cloud Bare Metal Server Setup

Why Bother with Bare Metal When You’ve Got VMs?

Let’s cut the cloud marketing fluff: bare metal isn’t for everyone. It’s for the folks who’ve stared blankly at ethtool -S eth0 output while their Kubernetes pod’s latency spikes like a caffeine-addicted squirrel. It’s for teams running high-frequency trading engines, GPU-accelerated AI training jobs that choke on virtualized PCIe overhead, or legacy database clusters that throw tantrums when they smell a hypervisor. Tencent Cloud’s Bare Metal Server (BMS) gives you dedicated physical hardware—no noisy neighbors, no vCPU contention, no ‘but it worked fine in dev!’ excuses. And yes, it’s surprisingly easy to set up—if you skip the documentation rabbit hole and know where the landmines are buried.

Step 1: Prerequisites — Less ‘Ready’, More ‘Don’t Panic Yet’

Before you click anything: grab your Tencent Cloud account (with billing verified—yes, they’ll check), enable the BMS service in the Service Activation console (it’s not auto-enabled, and no, the error message won’t tell you that), and ensure your VPC is already built with at least one subnet in the target AZ. Bonus points if you’ve pre-allocated an Elastic IP—but hold off assigning it until after provisioning. Why? Because Tencent assigns a temporary internal IP during setup, and binding an EIP too early can cause DHCP confusion. Also: double-check your security group rules. Default ones allow nothing. Not even ping. Your first SSH attempt will fail gloriously—and you’ll blame the OS installer. Don’t.

Step 2: Hardware Selection — Where ‘General Purpose’ Lies

In the BMS console, go to Bare Metal Server > Create Instance. The instance type selector looks like a cereal aisle: dozens of options, half with names like ‘SA3.M540’ or ‘EBM.GN10X.24’. Here’s the cheat sheet:

• SA series: Balanced compute + memory. Good for databases, web backends, CI runners. Avoid if you need GPUs or ultra-low-latency storage.
• EBM.GN series: GPU-optimized. Comes with NVIDIA A10/A100/T4. Requires separate GPU driver install post-boot—and yes, you must disable Nouveau manually. (We’ll get there.)
• EBM.HD series: High-density storage. 24+ NVMe drives. For object storage gateways or log aggregation stacks. Warning: boot volume is still on SATA SSD—don’t put your root FS there if you need speed.

Pro tip: Skip local disks unless you’re doing ephemeral caching. Tencent’s local disk I/O is fast but volatile—reboot = data gone. Use CBS (Cloud Block Storage) for persistence. And ignore ‘Enhanced Networking’ checkboxes—they’re enabled by default and silently ignored on BMS anyway.

Step 3: OS & Network — The ‘Wait, It’s Already Running?’ Moment

Select your OS. Ubuntu 22.04 LTS? Great. CentOS 7? Still supported—but end-of-life whispers grow louder daily. Tencent provides custom ISOs with their kernel modules baked in (especially for RDMA and NVMe drivers). Use them. Don’t upload your own minimal ISO unless you enjoy compiling tencent-bms-tools from GitHub at 2 a.m.

Network setup is deceptively simple: choose your VPC, subnet, and assign a private IP. Leave ‘Auto-assign Public IP’ unchecked. Why? Because Tencent’s public IP assignment for BMS uses SNAT behind the scenes—great for outbound, terrible for inbound services. Instead, create an EIP separately, then bind it after the server boots and passes health checks. Also: enable ‘Enable Network ACL’ only if you love debugging packet drops. Default ACLs are restrictive; start with ‘Allow All’ and tighten later.

Step 4: Deployment — From ‘Creating’ to ‘Root@bms-xyz:~#’ in Under 8 Minutes

Click ‘Create’. Watch the status bar. It’ll say ‘Creating’ for ~90 seconds, then ‘Installing OS’ for another 3–4 minutes. Don’t refresh. Don’t panic. This isn’t a VM—there’s real PXE booting, disk partitioning, and kernel init happening. When status flips to ‘Running’, wait 60 seconds more. Then try SSH.

Your login credentials? If you selected key-pair auth: use your private key. If password auth: check the ‘Instance Details’ page—Tencent auto-generates and displays it *once*, in plain text. Copy it before navigating away. Lost it? Reset via console—but that triggers a full OS reinstall. Yes, really.

Step 5: First Login & Immediate Housekeeping

Once in, run these immediately:

sudo apt update && sudo apt upgrade -y  # Ubuntu
sudo yum update -y                      # CentOS
sudo timedatectl set-timezone Asia/Shanghai
sudo systemctl enable unattended-upgrades  # Ubuntu (if you dare)

Then: disable IPv6 if unused (sysctl -w net.ipv6.conf.all.disable_ipv6=1), rotate SSH host keys (rm /etc/ssh/ssh_host_* && dpkg-reconfigure openssh-server), and verify NTP sync with timedatectl status. If it says ‘inactive’, run sudo systemctl restart systemd-timesyncd.

Step 6: GPU Drivers (If You Chose GN Series)

For NVIDIA GPUs, skip the CUDA toolkit installer. Use Tencent’s prebuilt drivers:

wget https://mirrors.tencent.com/tencent-cloud-driver/nvidia-driver-535.129.03.run
sudo chmod +x nvidia-driver-535.129.03.run
sudo ./nvidia-driver-535.129.03.run --silent --no-opengl-files

Then blacklist Nouveau before reboot:

echo 'blacklist nouveau' | sudo tee /etc/modprobe.d/blacklist-nouveau.conf
echo 'options nouveau modeset=0' | sudo tee -a /etc/modprobe.d/blacklist-nouveau.conf
sudo update-initramfs -u  # Ubuntu
sudo dracut --force       # CentOS
sudo reboot

After reboot: nvidia-smi should show your GPU and driver version. If it says ‘No devices found’, check dmesg | grep -i nvidia—usually a Nouveau conflict you missed.

Step 7: Post-Deployment Reality Checks

Run these sanity checks:

• Huawei Cloud Cashback Credits lshw -class cpu,memory,disk,network -short → confirm you see all physical cores, RAM, and NVMe drives.
• ip link show → verify your primary interface is eth0, not ens3 or enp1s0f0. Tencent standardizes on eth0 for consistency.
• smartctl -a /dev/nvme0n1 → check SSD health. Yes, even new ones can ship with bad blocks.
• ping -c 4 114.114.114.114 → test DNS-independent connectivity. If this fails but curl https://www.qq.com works, your DNS resolver is misconfigured.

Finally: set up monitoring. Tencent’s Cloud Monitor agent works out-of-the-box on BMS—but install it before adding load. It collects hardware metrics (fan speed, CPU temp, disk SMART) that VM agents can’t touch.

What NOT to Do (Learned the Hard Way)

- Don’t attach multiple CBS volumes and expect LVM to auto-assemble. Tencent doesn’t inject LVM configs. Initialize manually.
- Don’t use ‘Cloud Init’ user-data scripts for critical boot tasks. BMS supports it, but timing is flaky—network may not be ready when it runs.
- Don’t assume ‘high availability’ means automatic failover. BMS has no built-in HA. You need your own keepalived + VIP or application-level redundancy.
- And whatever you do—don’t rename the default root user. Tencent’s agent and console features assume it exists. Renaming breaks support ticket diagnostics.

Wrapping Up: Bare Metal Isn’t Magic—It’s Just Metal

Tencent Cloud’s BMS delivers what it promises: raw, unvirtualized hardware with cloud-like provisioning speed. Setup isn’t wizard-level simple, but it’s predictable—once you know which defaults lie and which buttons silently ignore your clicks. The real win isn’t speed (though yes, fio --name=randread --ioengine=libaio --rw=randread --bs=4k --numjobs=4 --size=1G --runtime=60 will make you grin). It’s control. Knowing every interrupt, every cache line, every thermal throttle is yours to monitor, tune, or curse at. So go ahead—provision that server. Just remember: the power button is metaphorical. The real one is in your terminal. And it’s always been there.