Optimizing Network Statistics Retrieval in OSHI

Introduction

Open-source contributions are a great way to improve existing software while learning best practices and real-world problem-solving. Recently, I had the opportunity to contribute to OSHI (Open Source Hardware Info), a widely used Java library that provides system information.

My contribution focused on optimizing the way network statistics are retrieved in OSHI, replacing a command based approach with a more direct and performant method.

This blog will walk through the approach, challenges, and problem-solving strategies that led to a successful optimization.

Issue: https://github.com/oshi/oshi/issues/2826

PR: https://github.com/oshi/oshi/pull/2846

Understanding the Problem

OSHI provides detailed system information, including CPU, memory, disk, and network statistics. Previously, network statistics for TCP and UDP were being fetched using external command execution (netstat).

This approach had several drawbacks:

1. Performance Overhead: Spawning a new process to execute a command and parsing its output adds unnecessary resource consumption.

2. Redundant System Calls: Every time OSHI needed network stats, it executed a new system call, which was inefficient.

The goal was to replace this approach with a more efficient, native method using Linux's built-in system files.

Identifying an Alternative Solution

Discovering /proc/net/snmp and /proc/net/snmp6

Linux provides detailed network statistics through the /proc/net/snmp and /proc/net/snmp6 files. These files contain counters for network protocols (IP, ICMP, TCP, and UDP) in a structured format.

A sample output of /proc/net/snmp looks like this:

Tcp: RtoAlgorithm RtoMin RtoMax MaxConn ActiveOpens PassiveOpens AttemptFails EstabResets CurrEstab InSegs OutSegs RetransSegs InErrs OutRsts InCsumErrors
Tcp: 1 200 120000 -1 11447 33 0 5 1 382240 269403 298 0 322 0
Udp: InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti
Udp: 48942 16 0 48975 0 0 0 0

This data is:
✅ Well-structured
✅ Easily readable
✅ Consistently formatted across Linux distributions

By reading these files directly, we can eliminate external command execution and improve performance significantly.

Approach to Optimizing the Implementation

Step 1: Efficient File Parsing

Since /proc/net/snmp contains multiple protocol statistics (TCP, UDP), we needed to extract only the relevant TCPv4 and UDPv4 statistics efficiently.

💡 Optimization:

• Scanned for specific headers (e.g., "Tcp:", "Udp:") to locate the required data.

• Used structured mapping (stringToEnumMap) to parse values directly into Java Enums.

Step 2: Handling Parsing Challenges

Issue 1: Incorrect Line Splitting

Initially, the file parsing function (parseByteArrayToStrings) assumed null (\0) termination, but /proc/net/snmp uses newlines (\n).

🔍 Issue: The entire file was treated as a single string, leading to incorrect processing.
✅ Fix: Modified the function to correctly split by newline (\n), ensuring each line was processed separately.

Step 3: Further Optimization for UDPv6

UDPv6 statistics are stored in /proc/net/snmp6, but unlike /proc/net/snmp, each stat is on a separate line:

Udp6InDatagrams                     0
Udp6NoPorts                         0
Udp6InErrors                        0
Udp6OutDatagrams                    27155

💡 Optimization:

• Instead of scanning top to bottom, we iterated bottom to top because required stats are near the bottom.

• Used a counter to stop early when all needed values were found.

• This reduced the number of iterations, making the lookup more efficient.

Refer Linux code → https://github.com/torvalds/linux/blob/master/net/ipv6/proc.c#L217

Results

✅ Eliminated external process execution (netstat), reducing system calls.
✅ Improved performance by reading directly from /proc/net/snmp.
✅ Optimized parsing for UDPv6 by reading bottom-up and breaking early.