Ceph Storage Cluster Cost Optimization ลดค่าใช้จ่าย

Ceph Cost Optimization

Ceph Storage Cluster Cost Optimization Erasure Coding Replication Tiering Compression HDD SSD NVMe CRUSH OSD RBD RGW CephFS Kubernetes OpenStack

Cluster Design

# === Ceph Cluster Design ===

# Hardware Recommendations
# OSD Nodes (Storage):
#   CPU: 2x Intel Xeon 16-core
#   RAM: 128GB (16GB per OSD recommended)
#   HDD: 12x 16TB SATA (for capacity)
#   SSD: 2x 1TB NVMe (for WAL+DB)
#   Network: 2x 25GbE (public + cluster)
#
# Monitor Nodes:
#   CPU: 8-core
#   RAM: 32GB
#   SSD: 500GB NVMe
#   Network: 10GbE
#   Count: 3 or 5 (odd number)
#
# Manager Nodes (co-located with MON):
#   Dashboard, Prometheus, Grafana

# ceph.conf
# [global]
# fsid = a1b2c3d4-e5f6-7890-abcd-ef1234567890
# mon_initial_members = mon1, mon2, mon3
# mon_host = 10.0.0.1, 10.0.0.2, 10.0.0.3
# public_network = 10.0.0.0/24
# cluster_network = 10.0.1.0/24
#
# [osd]
# osd_memory_target = 4294967296  # 4GB per OSD
# bluestore_cache_size = 3221225472  # 3GB
# osd_max_backfills = 2
# osd_recovery_max_active = 2

from dataclasses import dataclass

@dataclass
class ClusterNode:
    role: str
    count: int
    cpu_cores: int
    ram_gb: int
    storage: str
    network: str
    cost_per_node: float

nodes = [
    ClusterNode("OSD (HDD)", 6, 32, 128, "12x 16TB HDD + 2x 1TB NVMe", "2x 25GbE", 8000),
    ClusterNode("OSD (SSD)", 3, 16, 64, "8x 4TB NVMe", "2x 25GbE", 12000),
    ClusterNode("MON/MGR", 3, 8, 32, "500GB NVMe", "10GbE", 2000),
    ClusterNode("RGW (S3)", 2, 8, 32, "500GB NVMe", "25GbE", 2500),
]

print("=== Cluster Hardware ===")
total_cost = 0
for n in nodes:
    cost = n.cost_per_node * n.count
    total_cost += cost
    print(f"  [{n.role}] x{n.count}")
    print(f"    CPU: {n.cpu_cores}c | RAM: {n.ram_gb}GB | Storage: {n.storage}")
    print(f"    Network: {n.network} | Cost: ")
print(f"\n  Total Hardware: ")

Erasure Coding และ Compression

# === Erasure Coding & Compression ===

# Create EC Pool
# ceph osd erasure-code-profile set ec-4-2 \
#   k=4 m=2 \
#   crush-failure-domain=host
# ceph osd pool create ec-data 128 erasure ec-4-2
# ceph osd pool set ec-data allow_ec_overwrites true
#
# # Enable Compression
# ceph osd pool set ec-data compression_algorithm zstd
# ceph osd pool set ec-data compression_mode aggressive
# ceph osd pool set ec-data compression_required_ratio 0.875
#
# # Create Replicated Pool (for hot data)
# ceph osd pool create hot-data 64 replicated
# ceph osd pool set hot-data size 3 min_size 2
#
# # CRUSH Rules — Separate SSD and HDD
# ceph osd crush rule create-replicated ssd-rule default host ssd
# ceph osd crush rule create-replicated hdd-rule default host hdd
# ceph osd pool set hot-data crush_rule ssd-rule
# ceph osd pool set ec-data crush_rule hdd-rule

@dataclass
class StorageSaving:
    strategy: str
    raw_tb: float
    usable_tb: float
    savings_pct: float
    monthly_cost: float

# Example: 100TB raw data
savings = [
    StorageSaving("3x Replication (baseline)", 300, 100, 0, 3000),
    StorageSaving("2x Replication", 200, 100, 33, 2000),
    StorageSaving("EC k=4 m=2", 150, 100, 50, 1500),
    StorageSaving("EC k=4 m=2 + Compression", 105, 100, 65, 1050),
    StorageSaving("EC k=8 m=3 + Compression", 96, 100, 68, 960),
]

print("\n=== Cost Savings (100TB usable) ===")
for s in savings:
    print(f"  [{s.strategy}]")
    print(f"    Raw: {s.raw_tb}TB | Usable: {s.usable_tb}TB | Savings: {s.savings_pct}% | Cost: /mo")

Monitoring และ Capacity

# === Monitoring & Capacity Planning ===

# Ceph Dashboard
# ceph mgr module enable dashboard
# ceph dashboard create-self-signed-cert
# ceph dashboard ac-user-create admin -i /tmp/password administrator

# Prometheus Integration
# ceph mgr module enable prometheus
# # Scrape endpoint: http://mgr-host:9283/metrics

# Key Metrics to Monitor
# - ceph_osd_utilization (target < 75%)
# - ceph_pool_stored_raw (actual data)
# - ceph_osd_apply_latency_ms (target < 10ms SSD, < 50ms HDD)
# - ceph_osd_commit_latency_ms
# - ceph_pg_degraded (should be 0)
# - ceph_health_status (HEALTH_OK)

# Capacity Planning
# ceph df detail
# ceph osd df tree

@dataclass
class PoolMetric:
    pool: str
    type: str
    stored_tb: float
    raw_tb: float
    usage_pct: float
    iops: int
    latency_ms: float

pools = [
    PoolMetric("kubernetes-rbd", "3x Replica", 8.5, 25.5, 72, 15000, 2.5),
    PoolMetric("object-storage", "EC 4+2", 45.0, 67.5, 58, 3000, 8.0),
    PoolMetric("backup-archive", "EC 8+3", 120.0, 165.0, 65, 500, 25.0),
    PoolMetric("cephfs-data", "3x Replica", 5.0, 15.0, 45, 8000, 3.0),
    PoolMetric("vm-images", "2x Replica", 12.0, 24.0, 60, 5000, 5.0),
]

print("Pool Dashboard:")
total_stored = 0
total_raw = 0
for p in pools:
    total_stored += p.stored_tb
    total_raw += p.raw_tb
    print(f"  [{p.pool}] {p.type}")
    print(f"    Stored: {p.stored_tb}TB | Raw: {p.raw_tb}TB | Usage: {p.usage_pct}%")
    print(f"    IOPS: {p.iops:,} | Latency: {p.latency_ms}ms")

print(f"\n  Total Stored: {total_stored:.1f}TB | Total Raw: {total_raw:.1f}TB")
print(f"  Effective Ratio: {total_raw/total_stored:.2f}x")

# Cost Optimization Checklist
checklist = [
    "EC สำหรับ Object Storage และ Archive ลด 50%+",
    "Compression zstd สำหรับ Text/Log Data ลด 30-50%",
    "CRUSH Tiering: SSD Hot / HDD Cold",
    "PG Autoscaler เปิด ไม่ต้อง Manual",
    "OSD Memory Target ปรับตาม RAM ที่มี",
    "Network แยก Public/Cluster ลด Congestion",
    "Capacity Alert ตั้งที่ 75% วางแผนซื้อล่วงหน้า",
]

print(f"\n\nOptimization Checklist:")
for i, c in enumerate(checklist, 1):
    print(f"  {i}. {c}")

เคล็ดลับ

• EC: ใช้ Erasure Coding สำหรับ Cold Data ประหยัด 50%+
• Compression: เปิด zstd Compression ลดอีก 30%
• Tiering: SSD สำหรับ Hot Data HDD สำหรับ Cold Data
• Monitor: ดู OSD Utilization ไม่เกิน 75%
• Network: แยก Public/Cluster Network ใช้ 25GbE

การนำความรู้ไปประยุกต์ใช้งานจริง

แหล่งเรียนรู้ที่แนะนำ ได้แก่ Official Documentation ที่อัพเดทล่าสุดเสมอ Online Course จาก Coursera Udemy edX ช่อง YouTube คุณภาพทั้งไทยและอังกฤษ และ Community อย่าง Discord Reddit Stack Overflow ที่ช่วยแลกเปลี่ยนประสบการณ์กับนักพัฒนาทั่วโลก

สรุป

Ceph Storage Cluster Cost Optimization Erasure Coding Compression zstd Tiering SSD HDD CRUSH OSD Replication Capacity Planning Monitoring Kubernetes Production