Hey there, let's dig into your Major GC issue and walk through your current GC configuration to figure out actionable tweaks.

First, let's break down your existing flags and where potential gaps are:

• Fixed heap size (-Xms6144M -Xmx6144M): This is good practice—it avoids overhead from dynamic heap resizing, so we can keep this setting.
• SurvivorRatio=4: This sets the ratio of Eden to Survivor spaces as Eden:S0:S1 = 4:1:1. Assuming your JVM uses the default young generation size (roughly 1/3 of the total heap, ~2GB here), your Eden space is ~1.33GB, and each Survivor space is ~333MB. The core problem here is likely that Survivor spaces are too small to hold short-lived objects that survive Minor GC, forcing them to get promoted to the old generation prematurely. Over time, this fills up the old gen and triggers Major GC every few hours.
• No explicit GC collector specified: Your JVM is using the default Parallel GC (for Java 8 and earlier). While it's efficient for throughput, it doesn't handle premature promotion as gracefully as more adaptive collectors like G1.
• GC log settings: -XX:GCLogFileSize=128K is way too small—logs will rotate frequently, making it hard to track long-term GC trends and diagnose root causes.
• MaxPermSize=256m: If you're on Java 8 or later, PermGen is replaced by Metaspace, so this flag is deprecated and should be swapped for Metaspace-specific settings.

Let's go through targeted adjustments to reduce premature promotions and mitigate frequent Major GCs:

1. Optimize Young Generation & Survivor Spaces #

The core issue is objects being promoted to old gen too early—let's fix that:

• Fix young generation size: Explicitly set a larger young gen to give more room for short-lived objects. Aim for 30-40% of total heap (since your heap is 6GB, 2.4GB is a good starting point):

  -XX:NewSize=2457M -XX:MaxNewSize=2457M
  

• Adjust Survivor space utilization: Increase TargetSurvivorRatio (default 50%) to let Survivor spaces hold more live objects before promotion. This reduces unnecessary moves to old gen:

  -XX:TargetSurvivorRatio=80
  

• Tweak SurvivorRatio (optional): If you want a larger Eden space relative to Survivors, adjust SurvivorRatio to 8 (Eden:S0:S1=8:1:1) to balance Eden size and Survivor capacity:

  -XX:SurvivorRatio=8
  

• Control object tenuring: Use MaxTenuringThreshold to set how many Minor GCs an object can survive before promotion. Lowering it (e.g., to 10 from default 15) keeps objects in young gen longer if they're still short-lived—adjust based on your tenuring distribution logs later:

  -XX:MaxTenuringThreshold=10
  

• Add tenuring logs: To get visibility into object age distributions, add this flag to your config—it will show exactly how many objects are surviving each GC cycle, helping you fine-tune other settings:

  -XX:+PrintTenuringDistribution
  

2. Switch to a More Adaptive GC Collector (Recommended for Java 7+) #

G1 GC is designed to handle cases where premature promotion is an issue, as it dynamically adjusts young/old gen sizes and avoids long Major GC pauses. Here's how to configure it:

-XX:+UseG1GC
-XX:MaxGCPauseMillis=200  # Set your desired maximum pause time (adjust based on your app's latency needs)
-XX:InitiatingHeapOccupancyPercent=45  # Trigger G1's concurrent cycle earlier to avoid full GC

If you stick with Parallel GC, add this to optimize throughput and GC thread count:

-XX:ParallelGCThreads=8  # Match this to your CPU core count (adjust as needed)

3. Clean Up Deprecated & Log Settings #

• Replace PermGen with Metaspace (Java 8+):

  -XX:MetaspaceSize=256M -XX:MaxMetaspaceSize=256M
  

• Adjust GC log rotation: Increase log file size to avoid frequent rotations (100MB is a reasonable starting point):

  -XX:GCLogFileSize=100M
  

After applying these tweaks, monitor your gc.log closely to:

• Check if the number of objects promoted to old gen per Minor GC has decreased
• Track how often Major GC runs (it should become less frequent)
• Verify that Survivor spaces are being utilized effectively (use PrintTenuringDistribution data to fine-tune MaxTenuringThreshold and TargetSurvivorRatio)

内容的提问来源于stack exchange，提问作者Sankalp Gupta