VPS 67 • VPS Hosting

Direct admin support    30-day money-back guarantee    Cancel anytime

Intel Xeon E5-2680 v3    100 / 20 Mbit Connectivity    SAS 12 Gb Redundant RAID-Z1 array

Gainesville FL    KVM / Proxmox VE    Daily off-host Infiniband backups

Three plans built on Proxmox VE 9, RAID Z1 and ZFS — they run like a classic

VPS 67 • Two-Moto

$10 /mo

  • 2 vCPU Cores
  • 8 GB RAM
  • NVMe/Optane-boosted I/O
  • 400 GB SAS RAID-Z1
  • 100/20 Mbit/s
Order Now
VPS 67 • Inline-Four

$20 /mo

  • 4 vCPU Cores
  • 16 GB RAM
  • Optane SLOG + L2ARC caching
  • 800 GB SAS RAID-Z1
  • 100/20 Mbit/s
Order Now
VPS 67 • Eight-Banger

$40 /mo

  • 8 vCPU Cores
  • 32 GB RAM
  • Intel Optane SLOG + L2ARC
  • 1.6 TB SAS RAID-Z1
  • 100/20 Mbit/s
Order Now

Storage Performance & Architecture

VPS 67 runs on a tuned ZFS RAID-Z1 array backed by Intel Optane for both SLOG (write intent log) and L2ARC (read cache). This dramatically lowers latency and boosts 4K IOPS on virtual machines. Below are real measurements from this node.

SLOG Write Latency

Optane P4800X as dedicated ZFS SLOG
~88–90 µs avg latency
~10.8k IOPS (fsync=1)

L2ARC Read Cache

Optane L2ARC warm read throughput
~2.3–2.4 GB/s
sustained across ZFS pipeline

4K Random Reads

Real L2ARC 4K random read test
~124k IOPS
~7.6 µs avg latency

These results reflect the actual performance of the storage subsystem under Proxmox VE 9, with tuned ZFS parameters, SAS RAID-Z1 for capacity, and Optane for low-latency IO acceleration. Even under mixed VPS workloads, latency remains predictable and stable.

Real-world fio benchmarks

All numbers below are taken from fio runs on this VPS 67 node against the live ZFS RAID-Z1 pool with Intel Optane used as both SLOG and L2ARC. Values are rounded for readability.

Latency (lower is better)

SLOG 4K sync write (RAID-Z1 + Optane)

≈ 88–90 µs average fsync latency

4K random read from L2ARC (Optane)

≈ 7.6 µs average latency

IOPS & throughput

SLOG 4K sync writes (fsync=1)

≈ 10.6k IOPS sustained

4K random read from L2ARC

≈ 124k IOPS at ~4K block size

128K sequential read from L2ARC

≈ 2.3 GB/s sustained read throughput

Technical Specification • Storage Stack

ZFS Pool RAID-Z1 • 4 × 3.6 TB SAS (12 Gb/s)
SLOG Device Intel Optane P4800X (partitioned, 16 GB)
L2ARC Cache Intel Optane P4800X (~680 GB)
ARC (RAM Cache) 64 GB DRAM (ARC max: 64 GB, min: 32 GB)
Recordsize (VM) 16K (optimized for KVM / qemu)
Compression lz4 (realtime, CPU-efficient)
Special tuning logbias=latency, sync=always, xattr=sa, dnodesize=auto
Throughput (L2ARC) ~2.3 GB/s sequential read
IOPS (4K) ~124k IOPS random read (L2ARC)
Sync Write Latency ~88–90 µs (SLOG)

Measured under Proxmox VE 9 with ZFS 2.2 on Intel Xeon E5-2680 v3
Benchmark tools: fio and internal ARC/L2ARC counters.

VPS 67 Performance Whitepaper (Markdown)

Frequently Asked Questions

VPS 67 is a small-batch hosting project personally built and maintained by a single engineer — no outsourcing, no middlemen. Each node runs Proxmox VE 9 on enterprise-grade SAS drives with a ZFS RAID-Z1 array, tuned for reliability and transparent operation.

In addition, VPS 67 is powered by Intel Optane technology — a dedicated Optane SLOG and L2ARC caching layer that dramatically reduces write latency and boosts read performance compared to regular NVMe or SATA SSD-based hosts. This means more consistent I/O, lower jitter, and faster real-world performance even under load.

Support requests go directly to the person who built the system — you may use a ticket for convenience, but you’ll always be talking to a real admin, not a bot or a front-desk operator.

Backups are performed twice daily over a dedicated Infiniband link to a separate Proxmox Backup Server. Each backup is encrypted and stored on a redundant ZFS pool independent of the main node. To keep things efficient, no more than 10 most recent backups are retained per VPS, ensuring both data safety and optimal storage performance.

Yes. You can upgrade, downgrade, or cancel anytime. Plan changes are usually processed within 24 hours and prorated fairly. Just contact support — there are no hidden fees or penalties.

Absolutely. Each VPS comes with full root SSH access. You’re free to install, configure, and experiment as you wish — it’s your virtual server, your rules.

You can choose from common Linux distributions like Debian, Ubuntu, Rocky Linux, CentOS Stream. Custom ISOs are also supported — just send a link or upload your image for installation.

All VPS 67 nodes are located in Gainesville, Florida and operated in-house. This ensures minimal layers between hardware and user, low latency for regional users, and direct accountability for uptime and maintenance.

Each VPS is accessible through a unique subdomain such as vm_{uuid}.pve{n}.vms.sy1235.com, where {uuid} corresponds to your assigned virtual machine identifier and {n} denotes the physical node (bare metal server) running Proxmox VE. Connections are routed securely via SSH, HTTPS, or the Proxmox web console. All instances are fully reachable through their subdomain endpoints.

Outgoing connections on standard web and application ports such as 80 (HTTP) and 443 (HTTPS) are fully open. Certain mail-related ports — most notably 25, 465, and 587 — are currently restricted by firewall policy to protect network reputation and prevent abuse. Email delivery should instead be implemented through trusted providers using HTTPS APIs or other authenticated relay methods. Requests to enable outbound SMTP access for verified services such as Gmail, Mailgun, or Amazon SES can be reviewed on a case-by-case basis. No inbound port filtering is applied unless configured by the user’s own firewall rules.

All servers operate under continuous power protection. The facility is equipped with uninterruptible power supplies (UPS) and automated battery backup systems that maintain operation during utility outages and ensure a smooth transition when power is restored. In case of extended outages, the autonomous power system can sustain full operation for approximately 3 to 6 hours, depending on load conditions. Longer interruptions are considered force-majeure events.

Network connectivity is generally stable and continuously monitored. However, the upstream provider occasionally performs scheduled maintenance, typically during nighttime hours in the U.S. Eastern Time (ET) zone. Such events are infrequent — usually no more than once per month — and advance notifications are issued at least two weeks beforehand. During these maintenance windows, brief interruptions in connectivity may occur. Clients are notified in advance whenever maintenance affecting their services is planned.

Global Latency Comparison

Real-world latency measurements from the VPS 67 node in Gainesville, FL. Tests performed using mtr with 10-packet probes to major backbone providers (Tier-1 and mobile carriers). Values rounded for readability.

Provider / Region Avg Latency Rating Notes
AT&T Backbone ~41 ms Excellent Stable US East path
Verizon ~40–43 ms Excellent Direct via Atlanta
T-Mobile ~105 ms OK Heavy Zayo path (Dallas → LA)
Comcast ~21–24 ms Excellent Strong Atlanta POP
Charter / Spectrum ~38–52 ms Good Texas routing, stable
Google (AS15169) ~21–36 ms (US)
~120–133 ms (EU path)		 Excellent Direct peering in Atlanta
Telia (US Route) ~20–35 ms Excellent Atlanta backbone
Telia Europe (AS1299) Frankfurt: ~130 ms
London: ~125–130 ms
Amsterdam: ~20 ms*
Stockholm: ~20 ms* 		Excellent *ICMP filtered mid-path
Miami (Vultr) ~32–36 ms Excellent Clean 1299 → Miami route

Measurements taken on Proxmox VE 9 node via mtr -rw using 10 packets. Ratings are based on geographic expectations and backbone path quality.

Real Latency Benchmarks

Live ICMP latency measurements (mtr -rw) from the VPS 67 node in Gainesville, Florida.

AT&T • Southeast 41 ms
T-Mobile • US Transit 105 ms
Comcast / Xfinity 28 ms
Charter / Spectrum 45 ms
Google (USA Anycast) 21 ms
Verizon 24 ms
Zayo • Dallas → LA long-haul 108 ms
Miami (Vultr) 32 ms
Telia Europe • Frankfurt 130 ms
Telia Europe • London 125 ms
Telia Europe • Amsterdam 25–35 ms*

*ICMP filtering upstream — displayed hop is partial.

Telia Europe • Stockholm 25–35 ms*

*ICMP filtering upstream — displayed hop is partial.