Noida VPS Node Benchmark Results: CPU, Storage & Network Performance

In the cloud infrastructure industry, marketing materials are filled with grandiose claims: "lightning-fast processors," "ultra-low latency," and "blazing NVMe speeds." Yet, experienced software engineers and infrastructure architects know that marketing claims are meaningless without verifiable, reproducible telemetry.

When selecting an Indian cloud server for production web applications, database clusters, or latency-sensitive APIs, technical teams require concrete performance numbers: What are the raw multi-threaded CPU event speeds? What is the sustained random 4K read/write IOPS under queue depth 32? What is the actual round-trip network latency to major Indian transit corridors?

To provide complete transparency into our infrastructure, VisiHost conducted comprehensive, multi-hour hardware benchmarks on our production Noida VPS compute nodes. In this report, we publish the raw telemetry from industry-standard benchmarking suites—including Sysbench 1.0.20, FIO 3.36, and network routing telemetry across Delhi NCR and nationwide metros.

1. Test Node Environment & Hypervisor Architecture#

All tests documented in this report were executed on a standard, production-grade VisiHost Intel Xeon XS instance provisioned directly from our Noida, Uttar Pradesh data center facility.

Virtual Machine & Node Specifications

• Plan Tier: VisiHost Intel Xeon XS (₹389/month entry plan)
• Virtual CPU Allocation: 2 vCores (Intel Xeon Scalable Platform)
• Allocated Memory: 6,144 MB (6 GB) DDR4 Registered ECC RAM
• Storage Allocation: 50 GB Enterprise PCIe NVMe SSD (RAID array)
• Virtualization Layer: KVM (Kernel-based Virtual Machine) with virtio drivers
• Operating System: Ubuntu 24.04 LTS (Noble Numbat)
• Linux Kernel: 6.8.0-31-generic (x86_64)
• Network Uplink: 1 Gbps port with upstream transit routing

2. CPU Compute Benchmarks: Sysbench Multi-Threaded Processing#

To measure raw computational performance, memory bandwidth, and CPU event latency, we utilized Sysbench 1.0.20, the standard Linux system evaluation tool.

Test 1: CPU Prime Number Generation (Single-Thread vs Multi-Thread)

The Sysbench CPU test calculates prime numbers up to a maximum value of 20,000, measuring execution time, events per second, and latency distribution percentiles.

bash
COPY
# Sysbench CPU 2-Thread Test Command
sysbench cpu --cpu-max-prime=20000 --threads=2 run

Telemetry Results Summary

+-------------------------------------------------------------------------+
|                  SYSBENCH CPU BENCHMARK TELEMETRY                       |
+---------------------------------+-------------------+-------------------+
| Metric / Parameter              | 1 Thread (Single) | 2 Threads (Multi) |
+---------------------------------+-------------------+-------------------+
| Total Events Processed          | 10,000            | 10,000            |
| Total Execution Time            | 7.2148s           | 3.6421s           |
| Events Per Second               | 1,385.98 / sec    | 2,745.36 / sec    |
| Latency: Minimum                | 0.71 ms           | 0.70 ms           |
| Latency: Average                | 0.72 ms           | 0.73 ms           |
| Latency: 95th Percentile        | 0.73 ms           | 0.74 ms           |
| Latency: 99th Percentile        | 0.75 ms           | 0.76 ms           |
+---------------------------------+-------------------+-------------------+

Analysis

• Near-Linear Scaling: The 2-thread execution completed in 3.64 seconds compared to 7.21 seconds on a single thread—demonstrating a 99.2% multi-threaded scaling efficiency. This confirms that both assigned vCPU cores provide genuine compute capacity without artificial hypervisor CPU throttling.
• Latency Consistency: The difference between minimum latency (0.70 ms) and the 99th percentile (0.76 ms) is less than 0.06 ms. This near-zero jitter indicates that the underlying hypervisor node is not burdened by hypervisor overcommitment or resource contention from neighboring virtual machines.

3. Storage Subsystem Telemetry: FIO Enterprise NVMe Analysis#

Storage throughput and random I/O performance are the primary determinants of database responsiveness (PostgreSQL, MySQL, MongoDB) and application build speeds. We utilized FIO (Flexible I/O Tester) version 3.36 to evaluate our enterprise PCIe NVMe storage arrays.

Test 1: Random 4K Read/Write (Database Simulation)

Random 4K operations simulate the exact disk access patterns of relational databases executing concurrent queries, updating index trees, and appending to write-ahead logs (WAL).

bash
COPY
# FIO Random 4K Test Command (Direct I/O, Queue Depth 32, 4 Jobs)
fio --name=rand4k --ioengine=libaio --iodepth=32 --rw=randrw --rwmixread=75     --bs=4k --direct=1 --size=2G --numjobs=4 --runtime=60 --group_reporting

Telemetry Results: Random 4K Operations

+-------------------------------------------------------------------------+
|                  FIO RANDOM 4K READ/WRITE TELEMETRY                     |
+--------------------------+---------------------+------------------------+
| Metric / Parameter       | Read Performance    | Write Performance      |
+--------------------------+---------------------+------------------------+
| IOPS (IO Operations/Sec) | ~42,800 IOPS        | ~14,250 IOPS           |
| Bandwidth / Throughput   | 171.2 MB/s          | 57.0 MB/s              |
| Latency: Average         | 0.74 ms             | 2.24 ms                |
| Latency: 95th Percentile | 1.12 ms             | 3.85 ms                |
| Latency: 99th Percentile | 1.48 ms             | 4.92 ms                |
+--------------------------+---------------------+------------------------+
| Combined Total IOPS      |                57,050 IOPS                   |
+--------------------------+----------------------------------------------+

Test 2: Sequential 1MB Read/Write (Large File & Backup Simulation)

Sequential tests measure sustained bandwidth when streaming media, creating database dumps, or extracting large archives.

bash
COPY
# FIO Sequential 1MB Test Command
fio --name=seq1m --ioengine=libaio --iodepth=16 --rw=readwrite --bs=1M     --direct=1 --size=4G --numjobs=2 --runtime=30 --group_reporting

Telemetry Results: Sequential 1MB Operations

• Sequential Read Speed: 1,840 MB/s (1.84 GB/s)
• Sequential Write Speed: 925 MB/s
• Average Completion Latency: 8.65 ms

Analysis

Compared to budget hosting providers utilizing legacy SATA SSDs (which cap at ~500 MB/s sequential and 10,000–15,000 IOPS), VisiHost's enterprise PCIe NVMe storage array delivers 3.5x to 4x higher sustained disk throughput. For developers running PostgreSQL or MongoDB, this translates into zero disk bottlenecking during intensive database transactions.

4. Network Latency & Domestic Transit Telemetry#

Datacenter placement determines network packet transit times for users across the Indian subcontinent. To measure real-world network responsiveness, we conducted multi-point ICMP ping tests from our Noida compute node to major telecom nodes, regional gateways, and major cities across India.

+-------------------------------------------------------------------------+
|                  NOIDA VPS REGIONAL LATENCY BENCHMARK                   |
+--------------------------+---------------------+------------------------+
| Destination City / Hub   | Distance from Node  | Measured Round-Trip RTT|
+--------------------------+---------------------+------------------------+
| Noida / Greater Noida    | Local DC            | 1.8 - 2.8 ms           |
| New Delhi (Central)      | ~25 km              | 2.4 - 4.2 ms           |
| Gurugram (Haryana)       | ~45 km              | 3.2 - 5.6 ms           |
| Faridabad (Haryana)      | ~30 km              | 2.9 - 4.8 ms           |
| Chandigarh (Punjab)      | ~260 km             | 8.4 - 12.2 ms          |
| Jaipur (Rajasthan)       | ~290 km             | 9.8 - 14.5 ms          |
| Lucknow (UP East)        | ~510 km             | 10.2 - 15.8 ms         |
| Dehradun (Uttarakhand)   | ~240 km             | 7.8 - 11.4 ms          |
| Mumbai (Maharashtra)     | ~1,400 km           | 22.4 - 26.8 ms         |
| Pune (Maharashtra)       | ~1,450 km           | 24.2 - 28.5 ms         |
| Bengaluru (Karnataka)    | ~2,150 km           | 28.6 - 34.2 ms         |
| Hyderabad (Telangana)    | ~1,550 km           | 26.4 - 31.8 ms         |
| Kolkata (West Bengal)    | ~1,480 km           | 25.1 - 30.4 ms         |
| Chennai (Tamil Nadu)     | ~2,180 km           | 32.2 - 37.5 ms         |
+--------------------------+---------------------+------------------------+

Key Takeaways from Network Telemetry

1. The Delhi NCR Sub-6ms Corridor: For users and businesses in Delhi, Noida, Gurugram, and Faridabad, ping latency remains consistently under 6 ms. This makes Noida compute nodes ideal for local SaaS applications, payment gateways, and real-time WebSocket communication.
2. Comprehensive North India Coverage: Transit times to Uttar Pradesh, Punjab, Haryana, and Rajasthan remain firmly under 15 ms, providing lightning-fast page loads for over 400 million internet users in North India.
3. Sub-35ms Nationwide Reach: Upstream transit peering ensures that packets reach Southern and Western metros (Mumbai, Bengaluru, Hyderabad, Chennai) in under 35 ms, well below the 100 ms threshold where human users perceive interface lag.

4. Memory Subsystem Telemetry: Sysbench RAM Throughput#

Memory throughput directly influences high-concurrency Node.js garbage collection passes, in-memory caching layers (Redis, Memcached), and complex sorting algorithms in databases like PostgreSQL and MySQL.

Using Sysbench's memory benchmark suite, we evaluated both sequential and random memory transfer rates across our ECC Registered DDR4 memory channels:

bash
COPY
# Sysbench Sequential Memory Transfer Test Command
sysbench memory --memory-block-size=1M --memory-total-size=100G --memory-oper=write --memory-access-mode=seq run

Telemetry Results: Memory Bandwidth Analysis

+-------------------------------------------------------------------------+
|                  SYSBENCH MEMORY READ/WRITE TELEMETRY                   |
+--------------------------+---------------------+------------------------+
| Operation / Access Mode  | Total Data Tested   | Sustained Throughput   |
+--------------------------+---------------------+------------------------+
| Sequential Memory Write  | 100,000 MB (100 GB) | 16,842.15 MB/sec       |
| Sequential Memory Read   | 100,000 MB (100 GB) | 22,410.80 MB/sec       |
| Random Memory Write      | 50,000 MB (50 GB)   | 11,280.45 MB/sec       |
| Random Memory Read       | 50,000 MB (50 GB)   | 14,920.10 MB/sec       |
+--------------------------+---------------------+------------------------+

Analysis

• High-Throughput Bus: With sequential read speeds exceeding 22 GB/s and sequential write speeds nearing 17 GB/s, internal memory transfers execute without bandwidth saturation.
• ECC Stability: Error-Correcting Code (ECC) hardware registers prevent silent bit-flip memory corruption, providing enterprise stability for long-running processes, database buffers, and cryptographic services.

5. Network Throughput & iPerf3 Uplink Saturation#

While latency measures packet transit time, throughput measures bandwidth volume. To confirm that our 1 Gbps node ports deliver unthrottled bandwidth under heavy egress load, we ran synthetic iPerf3 bidirectional saturation tests against regional test endpoints:

bash
COPY
# iPerf3 10-Stream Parallel Uplink Test
iperf3 -c speedtest.noida.visihost.in -P 10 -t 30

Telemetry Results: Sustained Network Throughput

• Sustained Download Bandwidth: 942 Mbps (94.2% of physical 1 Gbps wire line rate).
• Sustained Upload Bandwidth: 918 Mbps (91.8% of physical line rate).
• Retransmitted Segments (TCP Retransmits): 0.003% under full wire saturation.
• Jitter under Maximum Load: 0.28 ms.

This confirms that VisiHost instances are connected to genuine gigabit switch fabric with unchoked upstream transit, capable of handling bursty web traffic spikes, large video streaming, or intensive database replication tasks without packet loss.

6. Thermal Stability & Multi-Hour Stress Testing#

Synthetic benchmarks that run for 30 seconds can be deceptive; some budget hosts configure aggressive "bursting" algorithms that deliver fast initial numbers but throttle CPU clocks as soon as thermal thresholds are reached.

To test sustained performance under continuous load, we subjected the test instance to a 4-hour continuous stress test utilizing stress-ng:

bash
COPY
# 4-Hour Sustained Stress Test Command
stress-ng --cpu 2 --vm 1 --vm-bytes 4G --timeout 4h --metrics-brief

Observations

• Zero Frequency Throttling: CPU core clock speeds remained completely stable throughout the 4-hour test window, showing zero thermal throttling or hypervisor eviction.
• Memory Integrity: Zero memory page faults, kernel allocation stalls, or OOM events were recorded across 4GB of continuously cycled physical RAM.
• Storage Reliability: Concurrent random I/O testing executed during the final 30 minutes of the stress test showed zero degradation in 4K read/write latency.

7. How to Reproduce These Benchmarks on Your Own Server#

We encourage developers to verify their hosting infrastructure independently. Below are the open-source commands you can run on any Linux VPS to test CPU, memory, storage, and network performance:

Step 1: Install Required Tools

bash
COPY
sudo apt update && sudo apt install -y sysbench fio curl

Step 2: Run Sysbench Multi-Core Test

bash
COPY
sysbench cpu --cpu-max-prime=20000 --threads=$(nproc) run

Step 3: Run FIO Random 4K Disk Test

bash
COPY
fio --name=fio-test --ioengine=libaio --iodepth=32 --rw=randrw     --rwmixread=75 --bs=4k --direct=1 --size=1G --numjobs=4     --runtime=30 --group_reporting

Step 4: Run Industry Standard Bench.sh Script

bash
COPY
curl -sL yabs.sh | bash

The popular YABS (Yet Another Bench Script) suite evaluates CPU performance with Geekbench, disk speed with FIO, and network throughput across global iPerf3 endpoints automatically.

8. Conclusion: Verified Performance at ₹389/Month#

Transparent infrastructure telemetry demonstrates what developers can expect from their hosting provider:

• Intel Xeon Scalable Compute: Consistent multi-threaded CPU throughput with zero noisy-neighbor jitter.
• Enterprise NVMe Storage: Over 57,000 combined 4K random IOPS and up to 1,840 MB/s sequential throughput for responsive databases.
• Noida Regional Latency: 2 to 6 ms ping across Delhi NCR and under 35 ms nationwide across all Indian metros.
• 100% KVM Virtualization: Dedicated memory and kernel autonomy with no multi-year lock-in contracts.

Ready to test this hardware performance for your own applications? Explore Noida VPS hardware and launch your 6GB KVM cloud instance on VisiHost starting at just ₹389/month with domestic UPI payments in INR.