diff --git a/catalog/redis-bench/README.md b/catalog/redis-bench/README.md
new file mode 100644
index 0000000..86e0bb2
--- /dev/null
+++ b/catalog/redis-bench/README.md
@@ -0,0 +1,164 @@
+# redis-bench — Redis Server/Client Benchmark Workload
+
+Multi-VM database benchmarking workload with a persistent Redis server and
+`redis-benchmark` client(s). The server stores data on a 10Gi persistent volume
+with AOF durability; clients generate configurable load and report throughput and
+latency percentiles.
+
+## Architecture
+
+```
+                  ┌──────────────────────────────┐
+                  │ K8s Service                  │
+                  │ virtwork-redis-bench:6379     │
+                  └──────────┬───────────────────┘
+                             │
+                     ┌───────▼────────┐
+                     │  Server (1 VM) │
+                     │  redis-server  │
+                     │  :6379         │
+                     │  /var/lib/redis │◄── 10Gi persistent volume
+                     └───────▲────────┘
+                             │ DNS
+               ┌─────────────┼─────────────┐
+               │             │             │
+          ┌────┴───┐    ┌────┴───┐    ┌────┴───┐
+          │Client 0│    │Client 1│    │Client N│
+          │ bench  │    │ bench  │    │ bench  │
+          └────────┘    └────────┘    └────────┘
+                        (scale with --vm-count)
+```
+
+## Parameters
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `num-clients` | int | `10` | Concurrent clients per benchmark run (`-c`) |
+| `num-requests` | int | `100000` | Total requests per iteration (`-n`) |
+| `pipeline-size` | int | `1` | Pipelining factor (`-P`); `1` = no pipelining |
+| `test-mode` | string | `set,get` | Redis operations to benchmark (`-t`) |
+| `data-size` | int | `256` | Value size in bytes (`-d`) |
+| `loop-delay` | int | `30` | Seconds between benchmark iterations |
+| `maxmemory` | string | `256mb` | Redis server memory limit (`--maxmemory`) |
+
+### Parameter details
+
+**num-clients** — Number of parallel connections per `redis-benchmark` invocation.
+Higher values increase throughput but also increase latency. Typical range: 1–100.
+
+**num-requests** — Total operations per iteration. More requests produce more stable
+latency percentiles. Use 100000+ for reliable p99 numbers.
+
+**pipeline-size** — Batches multiple requests per round-trip. Setting this to 16+
+dramatically increases throughput by reducing network overhead. Start at 1 for
+latency-focused testing.
+
+**test-mode** — Comma-separated list of Redis commands to benchmark. Options include
+`set`, `get`, `incr`, `lpush`, `lpop`, `sadd`, `spop`, `hset`, `hget`, `zadd`,
+`zrange`. Common combinations:
+
+- `set,get` — key-value read/write (default, most representative)
+- `get` — read-only (cache simulation)
+- `set` — write-only (ingest simulation)
+- `lpush,lpop` — list operations (queue simulation)
+
+**data-size** — Value payload in bytes for SET/GET. Larger values stress network I/O
+and memory. Typical range: 64–4096.
+
+**loop-delay** — Pause between benchmark runs. Allows distinct measurement windows for
+metrics scrapers.
+
+**maxmemory** — Server-side memory cap. When reached, Redis evicts keys using the
+`allkeys-lru` policy. Set relative to VM memory allocation.
+
+## Storage
+
+| Field | Value |
+|-------|-------|
+| Volume name | `redis-data` |
+| Size | `10Gi` |
+| Serial | `vw-redis` |
+| Device path | `/dev/disk/by-id/virtio-vw-redis` |
+| Mount point | `/var/lib/redis` |
+| Filesystem | XFS (auto-formatted) |
+
+The disk setup script (injected automatically by the catalog system) waits for the
+device, formats it with XFS if needed, mounts it, and adds an fstab entry. Redis
+AOF files are written to this volume for durability.
+
+## Usage
+
+```bash
+# Default: 10 clients, SET+GET, 100K requests per iteration
+virtwork run --from-catalog redis-bench --catalog-dir ./catalog --dry-run
+
+# High-throughput: 32 clients, pipelining, 1M requests
+virtwork run --from-catalog redis-bench --catalog-dir ./catalog \
+  --params redis-bench.num-clients=32,redis-bench.pipeline-size=16,redis-bench.num-requests=1000000
+
+# Read-heavy: GET-only, 64 clients, large values
+virtwork run --from-catalog redis-bench --catalog-dir ./catalog \
+  --params redis-bench.test-mode=get,redis-bench.num-clients=64,redis-bench.data-size=4096
+
+# Write-heavy: SET-only, small values
+virtwork run --from-catalog redis-bench --catalog-dir ./catalog \
+  --params redis-bench.test-mode=set,redis-bench.num-clients=50,redis-bench.data-size=64
+
+# Queue workload: list operations
+virtwork run --from-catalog redis-bench --catalog-dir ./catalog \
+  --params redis-bench.test-mode=lpush,redis-bench.num-clients=20
+
+# Scale to 3 client VMs
+virtwork run --from-catalog redis-bench --catalog-dir ./catalog \
+  --vm-count 3 --params redis-bench.num-clients=16,redis-bench.pipeline-size=8
+```
+
+## Monitoring
+
+### Server VM
+
+```bash
+# Check Redis is running
+systemctl status server.service
+journalctl -u server.service -f
+
+# Verify storage mount
+df -h /var/lib/redis
+ls -la /dev/disk/by-id/ | grep vw-redis
+
+# Redis stats
+redis-cli info server
+redis-cli info memory
+redis-cli info stats
+```
+
+### Client VM
+
+```bash
+# Check benchmark is running
+systemctl status client.service
+journalctl -u client.service -f
+```
+
+### Key metrics
+
+- **Throughput** (requests/sec) — typical range: 10K–100K ops/sec
+- **Latency** (avg/p50/p95/p99 in ms) — sub-millisecond for in-cluster traffic
+- **Memory usage** — server memory grows with data-size and working set
+- **Disk I/O** — AOF writes visible in storage metrics
+
+## Troubleshooting
+
+**Server fails to start** — Check that the storage volume mounted correctly.
+Inspect `/var/log/cloud-init-output.log` for disk setup errors. Verify the device
+exists: `ls /dev/disk/by-id/virtio-vw-redis`.
+
+**Client can't connect** — Verify the K8s Service exists (`oc get svc | grep redis`)
+and DNS resolves (`nslookup virtwork-redis-bench` from the client VM). Check that
+Redis is listening: `redis-cli -h <server-pod-ip> ping`.
+
+**Low throughput** — Increase `pipeline-size` to reduce round-trip overhead. Check
+for network contention with `oc top`. Ensure the server VM has enough CPU cores.
+
+**Redis OOM / evictions** — Increase `maxmemory` or reduce `data-size`. Monitor with
+`redis-cli info memory`. The `allkeys-lru` eviction policy is configured by default.
diff --git a/catalog/redis-bench/client.service b/catalog/redis-bench/client.service
new file mode 100644
index 0000000..f5e5ec1
--- /dev/null
+++ b/catalog/redis-bench/client.service
@@ -0,0 +1,12 @@
+[Unit]
+Description=Virtwork Redis benchmark client
+After=network-online.target
+
+[Service]
+Type=simple
+ExecStart=/bin/bash -c 'while true; do redis-benchmark -h virtwork-redis-bench -p 6379 -c {{num-clients}} -n {{num-requests}} -d {{data-size}} -P {{pipeline-size}} -t {{test-mode}} && sleep {{loop-delay}}; done'
+Restart=always
+RestartSec=10
+
+[Install]
+WantedBy=multi-user.target
diff --git a/catalog/redis-bench/server.service b/catalog/redis-bench/server.service
new file mode 100644
index 0000000..b58fde1
--- /dev/null
+++ b/catalog/redis-bench/server.service
@@ -0,0 +1,14 @@
+[Unit]
+Description=Virtwork Redis server with persistent storage
+After=network-online.target local-fs.target
+Requires=local-fs.target
+
+[Service]
+Type=simple
+ExecStartPre=/bin/bash -c 'mkdir -p /var/lib/redis && chown redis:redis /var/lib/redis'
+ExecStart=/usr/bin/redis-server --bind 0.0.0.0 --port 6379 --dir /var/lib/redis --appendonly yes --appendfilename appendonly.aof --maxmemory {{maxmemory}} --maxmemory-policy allkeys-lru
+Restart=always
+RestartSec=10
+
+[Install]
+WantedBy=multi-user.target
diff --git a/catalog/redis-bench/workload.yaml b/catalog/redis-bench/workload.yaml
new file mode 100644
index 0000000..eb660da
--- /dev/null
+++ b/catalog/redis-bench/workload.yaml
@@ -0,0 +1,53 @@
+description: "Redis server/client benchmark with persistent storage — multi-VM workload"
+
+packages:
+  - redis
+
+params:
+  - key: num-clients
+    type: int
+    default: "10"
+    desc: "Number of concurrent redis-benchmark clients (-c)"
+  - key: num-requests
+    type: int
+    default: "100000"
+    desc: "Total requests per benchmark iteration (-n)"
+  - key: pipeline-size
+    type: int
+    default: "1"
+    desc: "Pipelining factor — requests per batch (-P)"
+  - key: test-mode
+    type: string
+    default: "set,get"
+    desc: "Benchmark commands to run (-t), e.g. set,get,incr,lpush"
+  - key: data-size
+    type: int
+    default: "256"
+    desc: "Value size in bytes (-d)"
+  - key: loop-delay
+    type: int
+    default: "30"
+    desc: "Delay in seconds between benchmark iterations"
+  - key: maxmemory
+    type: string
+    default: "256mb"
+    desc: "Redis server maxmemory limit (e.g. 128mb, 1gb)"
+
+roles:
+  - name: server
+    vm-count: 1
+  - name: client
+    vm-count: 1
+
+storage:
+  - name: redis-data
+    size: 10Gi
+    serial: vw-redis
+    mount: /var/lib/redis
+
+service:
+  ports:
+    - name: redis
+      port: 6379
+      protocol: TCP
+  selector-role: server
diff --git a/catalog/sysbench/README.md b/catalog/sysbench/README.md
new file mode 100644
index 0000000..93484eb
--- /dev/null
+++ b/catalog/sysbench/README.md
@@ -0,0 +1,79 @@
+# sysbench — CPU Benchmarking Workload
+
+Single-VM continuous CPU stress test using [sysbench](https://github.com/akopytov/sysbench).
+Runs repeated benchmark iterations with configurable threads, CPU methods, and timing to
+generate sustained CPU load and throughput metrics.
+
+## Parameters
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `threads` | int | `4` | Number of worker threads (`--threads`) |
+| `cpu-method` | string | `sum` | CPU stressor method: `sum`, `sqrt`, or `pi` |
+| `test-duration` | int | `60` | Seconds per test iteration (`--time`) |
+| `loop-delay` | int | `10` | Seconds to pause between iterations |
+
+### Parameter details
+
+**threads** — Each thread performs independent CPU operations. Set this relative to VM
+CPU cores; values higher than the core count are valid and test contention behavior.
+
+**cpu-method** — Controls which mathematical operation each thread performs:
+
+- `sum` — integer addition (fastest, lowest per-op latency)
+- `sqrt` — square root calculations (moderate load)
+- `pi` — pi calculation via series (highest complexity per operation)
+
+**test-duration** — Duration of each benchmark pass. Use `0` for infinite (runs until
+stopped). Longer durations produce more stable throughput numbers.
+
+**loop-delay** — Pause between iterations. Allows metrics scraping systems to observe
+distinct test windows. Set to `0` for back-to-back runs.
+
+## Usage
+
+```bash
+# Default: 4 threads, sum method, 60s tests, 10s delay
+virtwork run --from-catalog sysbench --catalog-dir ./catalog --dry-run
+
+# High-intensity: 8 threads, pi method, 5-minute tests
+virtwork run --from-catalog sysbench --catalog-dir ./catalog \
+  --params sysbench.threads=8,sysbench.cpu-method=pi,sysbench.test-duration=300
+
+# Rapid iteration: 2 threads, 10s tests, 5s delay
+virtwork run --from-catalog sysbench --catalog-dir ./catalog \
+  --params sysbench.threads=2,sysbench.test-duration=10,sysbench.loop-delay=5
+
+# Cluster-wide stress: 4 VMs
+virtwork run --from-catalog sysbench --catalog-dir ./catalog \
+  --vm-count 4 --params sysbench.threads=4
+```
+
+## What it measures
+
+- **Throughput** (events/sec) — CPU operations completed per second
+- **Latency** (avg/p95/p99 in ms) — per-event timing distribution
+- **CPU utilization** — should approach 100% per thread under sustained load
+
+## Monitoring
+
+SSH into the VM and inspect the service:
+
+```bash
+systemctl status workload.service
+journalctl -u workload.service -f
+```
+
+sysbench prints a summary after each iteration with events/sec, latency
+percentiles, and total events processed.
+
+## Troubleshooting
+
+**Service fails immediately** — sysbench package not installed. Check
+`/var/log/cloud-init-output.log` for dnf errors.
+
+**Lower-than-expected throughput** — VM may be oversubscribed. Reduce `threads`
+or deploy on dedicated nodes.
+
+**CPU% below expected** — `test-duration` too short relative to `loop-delay`.
+Increase duration or decrease delay.
diff --git a/catalog/sysbench/workload.service b/catalog/sysbench/workload.service
new file mode 100644
index 0000000..466a0f3
--- /dev/null
+++ b/catalog/sysbench/workload.service
@@ -0,0 +1,12 @@
+[Unit]
+Description=Virtwork sysbench CPU benchmark
+After=network-online.target
+
+[Service]
+Type=simple
+ExecStart=/bin/bash -c 'while true; do sysbench cpu --threads={{threads}} --time={{test-duration}} --cpu-method={{cpu-method}} run; sleep {{loop-delay}}; done'
+Restart=always
+RestartSec=10
+
+[Install]
+WantedBy=multi-user.target
diff --git a/catalog/sysbench/workload.yaml b/catalog/sysbench/workload.yaml
new file mode 100644
index 0000000..de31fef
--- /dev/null
+++ b/catalog/sysbench/workload.yaml
@@ -0,0 +1,22 @@
+description: "Continuous CPU benchmarking with sysbench — configurable threads, methods, and duration"
+
+packages:
+  - sysbench
+
+params:
+  - key: threads
+    type: int
+    default: "4"
+    desc: "Number of sysbench worker threads (--threads)"
+  - key: cpu-method
+    type: string
+    default: "sum"
+    desc: "CPU test method: sum, sqrt, or pi (--cpu-method)"
+  - key: test-duration
+    type: int
+    default: "60"
+    desc: "Duration per test iteration in seconds (--time)"
+  - key: loop-delay
+    type: int
+    default: "10"
+    desc: "Delay in seconds between test iterations"
diff --git a/internal/workloads/catalog_entries_test.go b/internal/workloads/catalog_entries_test.go
new file mode 100644
index 0000000..fb8318f
--- /dev/null
+++ b/internal/workloads/catalog_entries_test.go
@@ -0,0 +1,105 @@
+// Copyright 2026 Red Hat
+// SPDX-License-Identifier: Apache-2.0
+
+package workloads_test
+
+import (
+	"os"
+	"path/filepath"
+	"runtime"
+
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+
+	"github.com/opdev/virtwork/internal/config"
+	"github.com/opdev/virtwork/internal/workloads"
+)
+
+func projectRoot() string {
+	_, file, _, _ := runtime.Caller(0)
+	return filepath.Join(filepath.Dir(file), "..", "..")
+}
+
+var _ = Describe("Shipped catalog entries", func() {
+	catalogDir := filepath.Join(projectRoot(), "catalog")
+
+	var entries []string
+
+	BeforeEach(func() {
+		dirEntries, err := os.ReadDir(catalogDir)
+		if os.IsNotExist(err) {
+			Skip("catalog directory not found at " + catalogDir)
+		}
+		Expect(err).NotTo(HaveOccurred())
+
+		entries = nil
+		for _, d := range dirEntries {
+			if d.IsDir() {
+				entries = append(entries, d.Name())
+			}
+		}
+		if len(entries) == 0 {
+			Skip("no catalog entries found in " + catalogDir)
+		}
+	})
+
+	It("should discover at least one entry", func() {
+		Expect(entries).NotTo(BeEmpty())
+	})
+
+	It("should load, validate, and produce working workloads for every entry", func() {
+		for _, name := range entries {
+			By("validating entry: " + name)
+
+			entry, err := workloads.LoadCatalogEntry(catalogDir, name)
+			Expect(err).NotTo(HaveOccurred(), "LoadCatalogEntry failed for %s", name)
+
+			schema := entry.Schema()
+			Expect(schema).NotTo(BeNil(), "Schema() returned nil for %s", name)
+
+			factory := entry.Factory()
+			Expect(factory).NotTo(BeNil(), "Factory() returned nil for %s", name)
+
+			cfg := config.WorkloadConfig{VMCount: 1, CPUCores: 2, Memory: "2Gi"}
+			opts := &workloads.RegistryOpts{
+				Namespace:   "test",
+				SSHUser:     "virtwork",
+				SSHPassword: "test",
+			}
+			wl := factory(cfg, opts)
+			Expect(wl).NotTo(BeNil(), "factory produced nil workload for %s", name)
+			Expect(wl.Name()).To(Equal(name), "Name() mismatch for %s", name)
+
+			userdata, err := wl.CloudInitUserdata()
+			Expect(err).NotTo(HaveOccurred(), "CloudInitUserdata failed for %s", name)
+			Expect(userdata).To(HavePrefix("#cloud-config"), "userdata missing #cloud-config prefix for %s", name)
+
+			if entry.IsMultiRole() {
+				multi, ok := wl.(workloads.MultiVMWorkload)
+				Expect(ok).To(BeTrue(), "multi-role entry %s did not produce MultiVMWorkload", name)
+
+				roles := multi.RoleDistribution()
+				Expect(roles).NotTo(BeEmpty(), "RoleDistribution empty for multi-role entry %s", name)
+
+				for _, rs := range roles {
+					roleUserdata, roleErr := multi.UserdataForRole(rs.Role, "test")
+					Expect(roleErr).NotTo(HaveOccurred(), "UserdataForRole(%s) failed for %s", rs.Role, name)
+					Expect(roleUserdata).To(
+						HavePrefix("#cloud-config"),
+						"role %s userdata missing #cloud-config for %s", rs.Role, name)
+				}
+			}
+
+			if len(entry.Manifest.Storage) > 0 {
+				dvts, dvtErr := wl.DataVolumeTemplates()
+				Expect(dvtErr).NotTo(HaveOccurred(), "DataVolumeTemplates failed for %s", name)
+				Expect(dvts).To(HaveLen(len(entry.Manifest.Storage)),
+					"DataVolumeTemplates count mismatch for %s: expected %d", name, len(entry.Manifest.Storage))
+			}
+
+			if entry.Manifest.Service != nil {
+				Expect(wl.RequiresService()).To(BeTrue(), "RequiresService should be true for %s", name)
+			}
+		}
+	})
+})