Compare The Io And Network Throughput Performance Of Different Brands Of 24-core Singapore Vps

2026-04-08 16:20:11
Current Location： Blog > Singapore VPS

1.

overview: comparison objectives and test focus

test objects: three common brands that provide 24-core vps in singapore (hereinafter referred to as a, b, and c).
test focus: disk io (sequential/random, 4k/1m, iops, latency) and network throughput (iperf3, packet loss, rtt).
test tools: fio 3.27, iperf3 3.10, ping, traceroute, and tcpdump are used for traffic skimming.
test environment: take the median of multiple measurements within the same time window to exclude short-term jitter and host hot spots.
purpose: to provide comparable performance data and selection suggestions for high-concurrency database, e-commerce and streaming media scenarios.

2.

test methods and configuration details

fio configuration example: fio --name=rand4k --rw=randread --bs=4k --size=10g --numjobs=8 --iodepth=32 --runtime=120 --group_reporting.
iperf3 test: two-way throughput test, both client and server use -p 8 multi-thread concurrency, test duration is 120s.
system level: kernel version linux 5.15, net.core.rmem_max=268435456, net.core.wmem_max=268435456, tcp_rmem/tcp_wmem is adjusted to automatically expand.
measurement items: sequential read and write, random 4k read and write (iops/latency), network peak bandwidth, round-trip delay, and packet loss rate.
sample collection method: each test is repeated 3 times, the median is taken, and the 95th percentile delay is recorded for sla reference.

3.

hardware and instance configuration examples

brand a (native nvme passthrough, sr-iov): 24vcpu (physically bound), 64gb ram, 2x1.6tb nvme, 10gbps capped bandwidth, kvm + sr-iov.
brand b (distributed ceph backend): 24vcpu (shared), 64gb ram, 1x1tb nvme (ceph rbd), 5gbps bandwidth, kvm + vswitch.
brand c (ordinary enterprise disk): 24vcpu (shared), 48gb ram, 2x1tb sata ssd (raid), 1gbps bandwidth, traditional virtualization such as xen/ovz.
software stack example: ubuntu 22.04, fio/iperf3, sysstat for monitoring, and cloud-init for basic network settings.
advantages of tuning: a enables cpu affinity and irq binding, b applies ceph client cache optimization, and c tries to configure the io scheduler to noop or mq-deadline.

4.

io performance test results and comparison

the test gives specific values ​​(median results in the table below, unit: mb/s or iops/ms):
brand sequential reading (1m) sequential writing (1m) random read (4k) random write (4k) 4k read latency (ms)
a (nvme pass-through) 2400mb/s 2100mb/s 200,000 iops 140,000 iops 0.8ms
b (ceph rbd) 1600mb/s 1200mb/s 90,000 iops 70,000 iops 2.1 ms
c (sata ssd) 500mb/s 420mb/s 25,000 iops 18,000 iops 8.5ms
conclusion: a is superior in random io and sequential throughput, and is suitable for highly concurrent database and log writing scenarios.
bdue to the distributed backend, latency and iops are affected by ceph osd and network, so it is suitable for disaster recovery and horizontal expansion needs.
c has the lowest cost but obviously limited io performance, making it suitable for static files or low-concurrency applications.

singapore vps

5.

network throughput and latency test results

use iperf3 -c server -p 8, two-way test, the results are as follows (median):
brand maximum bandwidth stable average bandwidth to hong kong rtt packet loss rate
a (10gbps port) 10 gbps 9.2 gbps 0.8ms 0.02%
b (5gbps port) 5 gbps 4.6 gbps 1.2 ms 0.1%
c (1gbps port) 1 gbps 930 mbps 2.5 ms 0.5%
it is observed that a is able to approach the line rate through sr-iov and pass-through nic, with low jitter and low packet loss.
b is due to the fact that virtualized switching and ceph traffic share a link with large periodic fluctuations.
c is limited by upstream bandwidth and is suitable for small and medium traffic loads.

6.

real case: comparison between e-commerce promotion and live broadcast scenarios

case 1 (e-commerce promotion): an e-commerce company deployed the main database and cache on a brand a 24-core vps. the peak order volume reached 8,000 requests per second, and the database was randomly written intensively. result: a can maintain low latency, 99% response is <30ms, and the promotion window is stable.
case 2 (live streaming): a live broadcast platform uses brand b as an edge forwarding node for evaluation. when live broadcast concurrency is high, network bandwidth competes with ceph write concurrency, resulting in short-term frame loss. subsequent adjustments were made to write the stream to the local ssd buffer and synchronize it to ceph asynchronously.
case 3 (small and medium-sized websites): choose brand c for static resource hosting, combine cdn (anycast) and cloud acceleration to reduce the bandwidth pressure on the origin site and resist small-scale ddos.
lesson: high io and high network concurrency usually cannot achieve low cost at the same time. a common approach is to separate hot and cold (local nvme cache + back-end distributed storage).
ddos defense: it is recommended to connect cloud cleaning (scrubbing) on ​​a/b or use cdn (caching static resources, anycast distribution) as the first line of defense.

7.

selection suggestions and optimization practices

if the target is a high-concurrency database or oltp: prioritize local nvme passthrough (such as brand a), and enable cpu affinity and irq binding.
if you are pursuing high availability and horizontal expansion: optional ceph backend (brand b), but the ceph cluster network and osd performance should be evaluated.
for cost-sensitive sites with low traffic: brand c combined with external cdn is a cost-effective combination.
network optimization: enable tcp window auto-tuning, adjust congestion control algorithms (such as bbr), use multi-thread concurrency (iperf3 -p), and enable sr-iov or dpdk to improve throughput.
security recommendations: access managed ddos protection, configure firewall speed limits, use anycast cdn to disperse traffic, and whitelist and rate limit traffic return sources for key businesses.

Tags：24-core vps singapore vps io performance network throughput iops iperf3 fio nvme ceph ddos defense cdn host comparison More»
Previous article： Optimization And Case Sharing Of Singapore’s Independent Cloud Servers Carrying High Concurrency During The E-commerce Promotion Period
Latest articles
Deployment Strategy For Offshore Cleaning Of Hong Kong High-Defense Servers In A Multi-Line Access Environment
Best Practices For Data Synchronization And DNS Switching During The Migration Of Native Vietnamese IP VPS
Key Compliance And Privacy Protection Considerations When Choosing Original IPs For Taiwan Services
Strategies For Negotiating Discounts On Bulk Purchases Of Korean Original IPs, Along With Recommendations For Long-term Maintenance Agreements
Bandwidth Optimization: How To Configure The Network Of Japanese Cloud Servers For Instant Response To Reduce Latency
Potential Service Risks And Assessment Checklist Behind The Low Prices Of High-security Servers In The United States
Comparison Of Latency Between Alibaba Cloud Hong Kong CN2 And Routes In Other Regions, Along With Selection Recommendations
Practical Tips: Use FIFA With A Hong Kong VPS To Connect To The US And Achieve Low-latency Multiplayer Gameplay
How To Set Up A Taiwan Proxy IP Server: Detailed Steps And Common Error Troubleshooting
An Operator’s Perspective On Why Alibaba Cloud Japan Doesn’t Use CN2 And An Assessment Of Its Impact On Access Speed
Popular tags
Overclocking Server Interaction Original Taiwan Ip Server Installation Ease Of Use Long-term Maintenance Telecom Routing Taiwan Registration Enhanced Safety Access Speed Taiwan Native Solid-state IP Service Network Interconnection Reasons For Selection Dota Taiwan Server Equipment Sharing File Sharing Stability Evaluation Security Website Building Pcidss Taiwan Proxies Traffic Attack Agency Purchasing Rental Tips Server Setup Infrastructure As Code Taiwan Server Vps Host Domain Name Cdnddos Operation And Maintenance Security Operational Efficiency Practical Recommendations Chroot
Related Articles