Skip to content
This repository was archived by the owner on Aug 9, 2026. It is now read-only.

WORKLOADS.md

Latest commit

 

History

History
180 lines (130 loc) · 4.53 KB

WORKLOADS.md

File metadata and controls

180 lines (130 loc) · 4.53 KB
Raw

Workloads

General

sudo apt install maven meson

To generate ZNS workloads run:

cd vendor/workloadgen/core
mvn -Dtest=site.ycsb.generator.TestZipfianGeneratorZNS test

To generate block-interface workloads run:

cd vendor/workloadgen/core
mvn -Dtest=site.ycsb.generator.TestZipfianGeneratorBLOCK test

Workloads will be located in target/workloadszoned (ZNS) and target/workloadsblock (block-interface).

THESE SCRIPTS WILL WIPE THE DISK!

To run ZNS workloads, run (replacing $DEVICE and $CONFIGFILE):

sudo  ./scripts/run_cpu_bench.sh vendor/workloadgen/core/target/workloadszoned $CONFIGFILE $DEVICE

To run block-interface workloads, run (replacing $DEVICE and $CONFIGFILE):

sudo  ./scripts/run_cpu_bench.sh vendor/workloadgen/core/target/workloadsblock $CONFIGFILE $DEVICE

Output will be in ./logs/* files

Eviction-threshold tuning

TestZipfianGeneratorZNSEvictTune generates workloads sweeping the eviction high/low-water (and clean-low) threshold matrix into target/workloads:

cd vendor/workloadgen/core
mvn -Dtest=site.ycsb.generator.TestZipfianGeneratorZNSEvictTune test

These are run with scripts/run_cpu_bench_evict.sh:

sudo ./scripts/run_cpu_bench_evict.sh vendor/workloadgen/core/target/workloads $NR_THREADS $CONFIGFILE $DEVICE

which runs each configuration in benchmark mode and appends throughput results to evict_bench.log.

WiredTiger trace workloads

The WiredTiger case study replays a block-access trace captured with the vendor/OxCache-WiredTiger-Trace submodule (an instrumented WiredTiger whose wtperf YCSB-C runner logs its block accesses). The trace is replayed in SPC format through the spclient binary via:

sudo ./scripts/run_cpu_bench_spc.sh $TRACE $CONFIGFILE $SPCCONFIGSPACE $DEVICE

with the eviction-parameter combinations ($SPCCONFIGSPACE) sourced from scripts/spcconfigspace.block.sh / scripts/spcconfigspace.zns.sh.

Cortes

On cortes, code is in /data/john/OxCache.

SSD: /dev/nvme1n1 ZNS: /dev/nvme0n2

Sanity check:

$ lsblk
NAME         MAJ:MIN RM   SIZE RO TYPE MOUNTPOINTS
sda            8:0    0 894.3G  0 disk
├─sda1         8:1    0   511M  0 part /boot/efi
├─sda2         8:2    0     1M  0 part
├─sda3         8:3    0 893.8G  0 part
│ ├─vg-swap  252:0    0     8G  0 lvm  [SWAP]
│ ├─vg-root1 252:1    0  92.7G  0 lvm  /
│ ├─vg-var1  252:2    0  10.5G  0 lvm  /var
│ ├─vg-root2 252:3    0  92.7G  0 lvm  /altroot
│ ├─vg-var2  252:4    0  10.5G  0 lvm  /altroot/var
│ └─vg-data  252:5    0 669.4G  0 lvm  /data
└─sda4         8:4    0     1M  0 part
nvme0n1      259:0    0     2G  0 disk
nvme0n2      259:1    0   1.8T  0 disk
nvme1n1      259:2    0 894.3G  0 disk

$ lsblk -z
NAME         ZONED
sda          none
├─sda1       none
├─sda2       none
├─sda3       none
│ ├─vg-swap  none
│ ├─vg-root1 none
│ ├─vg-var1  none
│ ├─vg-root2 none
│ ├─vg-var2  none
│ └─vg-data  none
└─sda4       none
nvme0n1      none
nvme0n2      host-managed
nvme1n1      none

Scheduler

IMPORTANT: On Linux older than 6.10 Must use a non re-ordering scheduler such as mq-deadline, set accordingly in /sys/block/$DEVICE/queue/scheduler

echo mq-deadline | tee /sys/block/$DEVICE/queue/scheduler
cat /sys/block/$DEVICE/queue/scheduler

Run simple:

sudo ./target/debug/oxcache --config cortes.server.zns.toml
sudo ./target/debug/evaluationclient --socket /tmp/oxcache.sock --num-clients 64 --data-file $FILE

Pre-conditioning

fio --name=precondition --filename=/dev/nvme1n1 --direct=1 \
    --rw=randwrite --bs=64k --size=100% --loops=2 \
    --randrepeat=0 --ioengine=libaio \
    --numjobs=1 --group_reporting

Graphing

Pre-processing

Data will be found in logs-compressed with each workload in a tar.gz

Move the data to $DATA_DIR

First extract:

for f in *.tar.gz; do dir="${f%%.tar*}"; mkdir -p -- "$dir"; tar --force-local -xzf "$f" -C "$dir"; done

Then consolidate which groups data:

python3 consolidate_metrics.py $DATA_DIR

This will create ${DATA_DIR}-consolidated

Repeat for ZNS and Block, putting data in separate dirs.

Then plot via:

./generate_all_plots.sh <zoned_dir>-consolidated <block_dir>-consolidated

See eval/README.md for the full figure/table reproduction guide, including the WiredTiger case-study pipeline (eval/generate_all_plots_wt.sh) and the WAF/RAF analysis.