GANESHA-QOS-CONFIG(8)

📦
version 9.16 · 4 versions
, in fedora 44

💖 NAME 💖

ganesha-qos-config — the NFS-Ganesha Quality of Service Configurations file ✨ (yes bestie, it's a section 8 man page, we're in the system-administration girlies now 💅)

📜 SYNOPSIS 📜

/etc/ganesha/ganesha.conf

✨ DESCRIPTION ✨

okay so. The nfs_qos module is literally part of the NFS-Ganesha server, and her whole job is making sure I/O resources get distributed fairly when things get busy 🫶. She does it with rate limiting techniques — bandwidth capping, IOPS (Input/Output Operations Per Second) management, and Token consumption. Obsessed.

The nfs_qos module lives in the source file nfs_qos.c 🧚. She's designed to manage quality of service for NFS operations by controlling how many operations get executed over a specified interval. Basically she balances "we want performance" against "please do not let the server get absolutely cooked by excessive or bursty I/O operations" 😭.

Here's how the module works, fr fr:

  • Bandwidth and IOPS Limiting: she monitors and regulates the amount of data transferred (bandwidth) and the number of ops executed (IOPS) 👑. And based on the configured limit's, she calculates the time required for the IO, and deffers the IO.
  • Token-Based Control: every I/O operation needs tokens to be processed 🎀. Tokens get replenished periodically. If an incoming request doesn't have enough tokens available, the IO will be retired slowly to the client with NFS_ERROR_DELAY 💀.
  • You can individually enable BandWidth, IOPS and Token controlling. Mix and match, main character energy ✨.

When you're deciding which QoS type to enable, here's the vibe check 🌈:

  • If simplicity and lower system overhead are the factors and the workload is fairly uniform, Per_Export might suffice. However — be aware that this can lead to client starvation 😭.
  • If your priority is making sure a client's queued I/Os don't indefinitely delay new requests from that same client, Per_Client may help, although it does not globally balance load across different exports.
  • If fairness at both the export and client level is paramount, and you can accommodate the additional complexity and resource usage, then Per_Export_Per_Client is the most balanced strategy 👑.

🌈 OPTIONS 🌈

The behavior of nfs_qos can be influenced by various configuration parameters, typically set in the server's main configuration file (for example, /etc/ganesha/ganesha.conf). Some configurable options include 💅:

  • Maximum bandwidth: set upper bounds on data transfer rates.
  • Maximun IOPS limits: set upper bounds for the number of operations executed per seconds.
  • Token thresholds and refresh intervals: define how many tokens available per time unit (usally 30 minutes, hourly, daily), which controls the rate of allowed data consuption over the specified time.

🎀 QOS_DEFAULT_CONFIG {} 🎀

The global defaults block, bestie. Every single knob, no skips, periodt 💖

  • enable_qos — bool, default false — she's off by default, so flip her on first 🌸
  • enable_tokens — bool, default false
  • enable_bw_control — bool, default false
  • enable_iops_control — bool, default false
  • enable_ds_control — bool, default false
  • combined_rw_bw_control — bool, default false
  • combined_rw_token_control — bool, default true ✨
  • combined_rw_iops_control — bool, default true ✨
  • qos_type — enum, values ["Per_Export", "Per_Client", "Per_Export_Per_Client"], default "Per_Export_Per_Client" 👑
  • max_export_combined_bw — uint64, range 32768 to 107374182400, default 2147483648
  • max_client_combined_bw — uint64, range 32768 to 107374182400, default 2147483648
  • max_export_write_bw — uint64, range 32768 to 107374182400, default 2147483648
  • max_export_read_bw — uint64, range 32768 to 107374182400, default 2147483648
  • max_client_write_bw — uint64, range 32768 to 107374182400, default 2147483648
  • max_client_read_bw — uint64, range 32768 to 107374182400, default 2147483648
  • max_export_iops — uint64, range 8 to 1638400, default 8192
  • max_client_iops — uint64, range 8 to 1638400, default 8192
  • max_export_write_iops — uint64, range 8 to 1638400, default 8192
  • max_export_read_iops — uint64, range 8 to 1638400, default 8192
  • max_client_write_iops — uint64, range 8 to 1638400, default 8192
  • max_client_read_iops — uint64, range 8 to 1638400, default 8192
  • max_export_write_tokens — uint64, range 3774873600 to UINT64_MAX, default 90596966400
  • max_export_read_tokens — uint64, range 3774873600 to UINT64_MAX, default 90596966400
  • max_client_write_tokens — uint64, range 3774873600 to UINT64_MAX, default 90596966400
  • max_client_read_tokens — uint64, range 3774873600 to UINT64_MAX, default 90596966400
  • export_write_tokens_renew_time — uint64, range 3600 to UINT64_MAX, default 86400
  • export_read_tokens_renew_time — uint64, range 3600 to UINT64_MAX, default 86400
  • client_write_tokens_renew_time — uint64, range 3600 to UINT64_MAX, default 86400
  • client_read_tokens_renew_time — uint64, range 3600 to UINT64_MAX, default 86400

🦋 EXPORT { QOS_BLOCK {} } 🦋

See global QOS_DEFAULT_CONFIG {} — same energy, different scope 💜

  • Config exception is qos_type is not available at export level 🙈
  • By default global EXPORT_DEFAULTS will be applied to export.
  • This block is required at export level only if there needs to be different value than global value.
  • Respective QOS_DEFAULT_CONFIG enable values needs to be enabled first to enable export level values.
  • If export QOS_BLOCK {} needs to be populated, then populate required enabled entities otherwise default value will get applied.
  • How range and default values derived can be found in src/include/nfs_qos.h 🧚

🌸 PSEUDOFS { QOS_BLOCK {} } 🌸

See EXPORT { QOS_BLOCK {} } — literally the same, bestie, go read that one ⭐

📝 NOTES:

The nfs_qos module is essential for maintaining a balanced and fair I/O workload on the NFS server 👑. By regulating operations with tokens, bandwidth, and IOPS constraints, she protects the server from overload while ensuring that client requests are handled in a controlled manner. This man page provides an overview of her function and operation within the greater context of the NFS server's architecture 🫶

👑 AUTHOR

The NFS-Ganesha-QOS feature was added by <Deeraj.Patil@ibm.com>, <Naresh.Chillarege@ibm.com> — legends, honestly 💅

💫 SEE ALSO

ganesha-config <>(8) 🌈

and that's the whole page, bestie!! this one came from NFS-Ganesha, dated Aug 11, 2026 — go forth and rate-limit responsibly, you absolute icon 💖✨🎀

📦 where this page comes from