TL;DR/summary:
- Tuning vm.watermark_boost_factor to 0 (disable) on Debian significantly improves performance on memory-intensive tasks that utilise SWAP space, by stopping preemptive kswapd freeing of memory, and subsequent page thrashing.
- I suggest that Debian should tune vm-watermark_boost_fact=0 by default to prevent this problem.
I have recently installed Debian 11 on a HP Z8 G4 Workstation (Z3Z16AV) - 32GB RAM, installed with ~120GB SWAP on a 2TB solid state drive (specs at end of this message).
I have been running some compute-intensive image processing tasks (CPU- and memory- intensive), which has on occasion had to dip into SWAP space, depending on image sizes (the processing I am running is image registration using elastix/transformix).
I had benchmarked the code on my Ubuntu laptop (similar spec) without any problems, but when running on Debian, whenever SWAP was needed, the system processing significantly slowed down/essentially froze.
After much debugging, I have traced this to the vm.watermark_boost_factor kernel parameter:
Comparing the Ubuntu and Debian kernel parameters using
sudo sysctl -a showed two key differences in virtual memory (vm) management parameters.
- Ubuntu:
- vm.swappiness=60
- vm.watermark_boost_factor=0
- Debian:
- vm.swappiness=10
- vm.watermark_boost_factor=150
I identified what these two parameters control:
I changed swappiness and then watermark_boost_factor sequentially, to see whether tuning these parameters to match my Ubuntu system prevented the system from freezing under my memory-intensive task.
- sudo sysctl vm.swappiness=60 on my Debian system did not prevent the freezing behaviour.
- sudo sysctl vm.watermark_boost_factor=0 (disabling it) on my Debian system prevented the freezing behaviour.
I then set these permanently by adding the following to /etc/sysctl.conf
vm.swappiness=60
vm.watermark_boost_factor=0
Further searching revealed this Ubuntu bug report:
swap storms kills interactive use
With this key entry:
Sultan Alsawaf (kerneltoast) wrote on 2020-03-27: #56
This problem is caused by an upstream memory management feature called watermark boosting. Normally, when a memory allocation fails and falls back to the page allocator, the page allocator will wake up kswapd to free up pages in order to make the memory allocation succeed. kswapd tries to free memory until it reaches a minimum amount of memory for each memory zone called the high watermark.
What watermark boosting does is try to preemptively fire up kswapd to free memory when there hasn't been an allocation failure. It does this by increasing kswapd's high watermark goal and then firing up kswapd. The reason why this causes freezes is because, with the increased high watermark goal, kswapd will steal memory from processes that need it in order to make forward progress. These processes will, in turn, try to allocate memory again, which will cause kswapd to steal necessary pages from those processes again, in a positive feedback loop known as page thrashing. When page thrashing occurs, your system is essentially livelocked until the necessary forward progress can be made to stop processes from trying to continuously allocate memory and trigger kswapd to steal it back.
This problem already occurs with kswapd *without* watermark boosting, but it's usually only encountered on machines with a small amount of memory and/or a slow CPU. Watermark boosting just makes the existing problem worse enough to notice on higher spec'd machines.
To fix the issue in this bug, watermark boosting can be disabled with the following:
# echo 0 > /proc/sys/vm/watermark_boost_factor
There's really no harm in doing so, because watermark boosting is an inherently broken feature...
So essentially, disabling watermark_boost_factor ensures effective swapping and reduces page thrashing.
I therefore suggest that Debian should tune vm.watermark_boost_factor=0 by default.
Cheers,
Steve.
Below are some more detailed specs of my Debian machine for reference:
$ uname -a
Linux panseer 5.10.0-11-amd64 #1 SMP Debian 5.10.92-1 (2022-01-18) x86_64 GNU/Linux
$ lscpu
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
Address sizes: 46 bits physical, 48 bits virtual
CPU(s): 20
On-line CPU(s) list: 0-19
Thread(s) per core: 2
Core(s) per socket: 10
Socket(s): 1
NUMA node(s): 1
Vendor ID: GenuineIntel
CPU family: 6
Model: 85
Model name: Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz
Stepping: 7
CPU MHz: 2511.149
CPU max MHz: 3200.0000
CPU min MHz: 1000.0000
BogoMIPS: 4800.00
Virtualization: VT-x
L1d cache: 320 KiB
L1i cache: 320 KiB
L2 cache: 10 MiB
L3 cache: 13.8 MiB
NUMA node0 CPU(s): 0-19
Vulnerability Itlb multihit: KVM: Mitigation: VMX disabled
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced IBRS, IBPB conditional, RSB filling
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Mitigation; TSX disabled
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon
pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic mov
be popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin ssbd mba ibrs ibpb stibp ibrs_enhanced tp
r_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd a
vx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req pku ospke avx512_
vnni md_clear flush_l1d arch_capabilities
$ free -h
total used free shared buff/cache available
Mem: 31Gi 3.6Gi 24Gi 160Mi 3.2Gi 26Gi
Swap: 119Gi 242Mi 118Gi
Steven J. West
BSc DPhil FRMS
_________________________________
International Brain Lab Histology Research Fellow
Sainsbury Wellcome Centre for Neural Circuits and Behaviour
University College London
25 Howland St, Fitzrovia, London W1T 4JG