Alibaba Cloud Linux extends the upstream Linux kernel with features that improve memory management, network performance, storage I/O, observability, and scheduling. This page lists every kernel extension by category, with the minimum supported kernel version and a description of what each feature does and why it exists.
Audience: System engineers and platform engineers who are familiar with the Linux kernel and want to identify which Alibaba Cloud Linux extensions apply to their environment.
Quick navigation: Memory | Network | Storage | Monitoring | Scheduling | Compatibility | Others
Memory
| Feature | Supported version | Description |
|---|
| Memcg Exstat feature | Alibaba Cloud Linux 2: 4.19.91-18.al7 and later<br>Alibaba Cloud Linux 3: all versions | The upstream kernel does not expose memory.events, memory.events.local, or memory.stat on the cgroup v1 interface. Alibaba Cloud Linux adds these three interfaces and provides latency statistics for memcg global minimum watermark changes and backend asynchronous reclamation. |
| Memcg global minimum watermark tiers | Alibaba Cloud Linux 2: 4.19.91-18.al7 and later<br>Alibaba Cloud Linux 3: all versions | Raises the global minimum watermark for resource-intensive tasks so they enter direct reclamation sooner, and lowers it for latency-sensitive tasks to protect them from direct reclamation. This prevents sudden memory allocations from causing performance degradation for latency-sensitive services. |
| Memcg QoS feature of the cgroup v1 interface | Alibaba Cloud Linux 2: 4.19.91-18.al7 and later<br>Alibaba Cloud Linux 3: all versions | The upstream kernel supports memcg quality of service (QoS) only on the cgroup v2 interface. Alibaba Cloud Linux also exposes memcg QoS on cgroup v1, letting you set memory usage locks and limits without migrating to cgroup v2. |
| Memcg backend asynchronous reclaim | Alibaba Cloud Linux 2: 4.19.81-17.al7 and later<br>Alibaba Cloud Linux 3: all versions | Provides per-memcg background memory reclamation using the workqueue mechanism instead of a dedicated kswapd thread, avoiding the overhead of an additional kernel thread per memcg. |
| Memcg OOM priority policy | Alibaba Cloud Linux 2: 4.19.91-18.al7 and later<br>Alibaba Cloud Linux 3: 5.10.60-9.al8 and later | When an out-of-memory (OOM) event occurs, selects the lowest-priority cgroup as the OOM kill target instead of relying solely on the default per-process oom_score_adj heuristic. |
| THP reclaim | Alibaba Cloud Linux 2: 4.19.91-24.al7 and later<br>Alibaba Cloud Linux 3: all versions | Resolves memory pressure and OOM errors caused by Transparent Huge Pages (THP) by enabling targeted THP reclamation. |
| Huge Pages | Alibaba Cloud Linux 2: 4.19.91-25 and later<br>Alibaba Cloud Linux 3: 5.10.112-11 and later | Maps executable code segments and dynamic-link libraries (DLLs) to huge pages to reduce instruction translation lookaside buffer (iTLB) misses and increase 2 MB iTLB utilization, improving application performance for large-code-segment workloads. |
| Kernel Electric-Fence (KFENCE) | Alibaba Cloud Linux 3: 5.10.84-10 and later (x86); 5.10.134-16 and later (Arm) | Enables dynamic detection of kernel memory pollution. Alibaba Cloud Linux supports enabling and disabling KFENCE at runtime in a flexible manner, fully capturing memory pollution issues and balancing online detection with offline debugging. |
| Page Cache Limit | Alibaba Cloud Linux 3: 5.10.134-14 and later | Caps page cache growth to prevent system instability—such as business jitters and unexpected OOM errors—caused by unbounded page cache usage. |
| Code duplication feature | Alibaba Cloud Linux 3: 5.10.112-11 and later | On non-uniform memory access (NUMA) systems—especially Arm-based instances—accessing code on a remote NUMA node adds latency. This feature replicates code segments from remote nodes into local node memory, eliminating cross-node code fetches. |
Network
| Feature | Supported version | Description |
|---|
| SMC-R | Alibaba Cloud Linux 3: 5.10.60-9.al8.x86_64 and later | Shared Memory Communications over RDMA (SMC-R) uses Alibaba Cloud elastic Remote Direct Memory Access (eRDMA) to replace TCP transparently, without changing application code. Provides direct memory-to-memory communication with lower latency than TCP, beneficial for in-memory databases, remote procedure calls (RPCs), and large file transfers. |
| TCP-RT configurations | Alibaba Cloud Linux 2: 4.19.91-21.al7 and later<br>Alibaba Cloud Linux 3: all versions | A TRACE-based method for monitoring TCP-layer service behavior. Tracks individual request-response pairs on a single connection and reports metrics such as kernel receive time and processing time. Also generates periodic per-connection statistics in the kernel. |
| TCP TIME-WAIT timeout | Alibaba Cloud Linux 2: 4.19.43-13.al7 and later<br>Alibaba Cloud Linux 3: all versions | The upstream Linux kernel fixes the TIME-WAIT duration at 60 seconds with no way to change it. Alibaba Cloud Linux exposes a kernel interface to shorten this timeout, improving throughput in high-connection-rate workloads. |
| Disable IPVS estimation | Alibaba Cloud Linux 2: 4.19.91-22 and later<br>Alibaba Cloud Linux 3: 5.10.134-14 and later | The IP Virtual Server (IPVS) estimation feature collects packet and connection statistics at regular intervals. On servers hosting many services—such as large Kubernetes clusters—this can introduce latency spikes of tens to hundreds of milliseconds. Disabling estimation eliminates this overhead when you don't need those statistics. |
| VTOA | Alibaba Cloud Linux 3: 5.10.134-15 and later | In FULLNAT scenarios, such as when Anti-DDoS Proxy is in use, the client address is rewritten to the FULLNAT node address. Install VTOA on the backend server and call getsockopt or getpeername using the TCP Option Address (TOA) protocol to retrieve the original client address. Supports both IPv4 and IPv6. |
Storage
| Feature | Supported version | Description |
|---|
| Cgroup writeback | Alibaba Cloud Linux 2: 4.19.36-12.al7 and later<br>Alibaba Cloud Linux 3: all versions | The upstream cgroup v1 interface cannot attribute dirty page writeback to the originating cgroup. Alibaba Cloud Linux backports cgroup writeback to cgroup v1, enabling per-cgroup buffered I/O rate limiting. |
| Blk-iocost weight-based throttling | Alibaba Cloud Linux 2: 4.19.81-17.al7 and later<br>Alibaba Cloud Linux 3: all versions | Extends the blkcg weight-based disk throttling model with a cost-model-driven controller (blk-iocost) that accounts for actual I/O cost rather than simple request counts, improving fairness under mixed workloads. |
| Block I/O throttling monitoring | Alibaba Cloud Linux 2: 4.19.81-17.al7 and later<br>Alibaba Cloud Linux 3: all versions | Adds kernel interfaces to expose richer block I/O throttling metrics, making it easier to diagnose throttling behavior under production workloads. |
| JBD2 optimization interface | Alibaba Cloud Linux 2: 4.19.81-17.al7 and later<br>Alibaba Cloud Linux 3: all versions | JBD2 is the journaling block device layer for ext4. Under certain conditions it enters a BH_Shadow state that degrades I/O performance. Alibaba Cloud Linux provides interfaces to work around this state and maintain consistent ext4 I/O throughput. |
| Hard links across project quotas | Alibaba Cloud Linux 2: 4.19.91-18.al7 and later<br>Alibaba Cloud Linux 3: all versions | By default, ext4 prevents hard links from being created across project quota boundaries. Alibaba Cloud Linux provides a custom interface to override this restriction when cross-quota hard links are required. |
| I/O latency tracking | Alibaba Cloud Linux 2: all versions<br>Alibaba Cloud Linux 3: all versions | Extends /proc/diskstats to include per-device read, write, and discard (special I/O) durations, providing richer data for the iostat tool. Also ships a BPF Compiler Collection (BCC) toolset for in-depth I/O latency analysis. |
| I/O hang detection | Alibaba Cloud Linux 2: all versions<br>Alibaba Cloud Linux 3: all versions | Extends core kernel data structures to detect and locate I/O hangs in file systems and block layers at low overhead, enabling faster diagnosis when a system stalls due to a slow I/O request. |
| Context readahead | Alibaba Cloud Linux 2: 4.19.91-18 and later<br>Alibaba Cloud Linux 3: all versions | The Linux context readahead algorithm detects interleaved sequential streams to prefetch data into the page cache. In random access workloads, it can overshoot and fetch unnecessary pages. Alibaba Cloud Linux lets you disable context readahead per file system to recover performance in those scenarios. |
| Tcm_loop device parameters | Alibaba Cloud Linux 3: 5.10.134-17 and later | Target Core Module (TCM) is the Linux-IO Target (LIO) in-kernel iSCSI target. TCM in Userspace (TCMU) lets userspace programs implement custom storage backends. Alibaba Cloud Linux exposes configurable parameters for the tcm_loop loopback module, simplifying userspace iSCSI target development. |
Monitoring
| Feature | Supported version | Description |
|---|
| UKFEF | Alibaba Cloud Linux 3: 5.10.60-9.al8 and later | The Unified Kernel Fault Event Framework (UKFEF) collects kernel exception events that may indicate system risk and emits them in a unified report format, making it easier to correlate fault events across subsystems. |
| Pressure Stall Information (PSI) for cgroup v1 | Alibaba Cloud Linux 2: 4.19.81-17.al7 and later<br>Alibaba Cloud Linux 3: all versions | The upstream kernel supports Pressure Stall Information (PSI) only on cgroup v2. Alibaba Cloud Linux backports PSI to cgroup v1, enabling per-cgroup CPU, memory, and I/O pressure monitoring without migrating to cgroup v2. |
Scheduling
| Feature | Supported version | Description |
|---|
| CPU burst for cgroup v1 | Alibaba Cloud Linux 2: 4.19.91-22.al7 and later<br>Alibaba Cloud Linux 3: all versions | The upstream cgroup v2 CPU controller supports CPU burst, allowing throttled containers to use accumulated slack bandwidth. Alibaba Cloud Linux backports this to cgroup v1, letting CPU-throttled containers burst above their quota to reduce tail latency. |
| Group identity | Alibaba Cloud Linux 2: 4.19.91-24.al7 and later<br>Alibaba Cloud Linux 3: 5.10.46-7.al8 and later | Assigns scheduling identity labels to CPU cgroups so the kernel can prioritize or deprioritize groups of processes, enabling more predictable latency for high-priority workloads running alongside batch jobs. |
Compatibility
| Feature | Supported version | Description |
|---|
| Anonymous pages statistics change in /proc/meminfo | Alibaba Cloud Linux 2: 4.19.91-22.al7 and later<br>Alibaba Cloud Linux 3: all versions | Starting from kernel 4.19.91-22.al7, the Active(anon) and Inactive(anon) counters in /proc/meminfo are calculated differently. If your application reads these values, data collected before and after upgrading to this kernel version will be inconsistent. |
Others
| Feature | Supported version | Description |
|---|
| Container resource visibility | Alibaba Cloud Linux 2: 4.19.91-22.2.al7 and later<br>Alibaba Cloud Linux 3: 5.10.46-7.al8 and later | Provides kernel interfaces that expose container resource usage, making it possible for monitoring tools inside containers to see accurate resource consumption rather than host-level totals. |
| User namespace security hardening | Alibaba Cloud Linux 2: all versions<br>Alibaba Cloud Linux 3: all versions | Linux user namespaces let processes assume different UIDs and GIDs across namespace boundaries, but this capability increases the attack surface for privilege escalation exploits. Alibaba Cloud Linux applies additional restrictions to user namespaces to reduce this risk. |