When configuring Linux virtual machines on VMware ESX/ESXi hosts, administrators often debate whether to allocate swap space within the guest OS or rely on hypervisor-level memory management. Let's examine the technical implications of both approaches through concrete examples.

Consider these two test scenarios:

# Scenario 1: Guest with swap partition
VM Configuration:
- 1GB RAM allocation
- 1GB swap partition
- Application consumes 1.5GB memory

# Scenario 2: Guest without swap
VM Configuration:
- 1GB RAM allocation
- No swap partition
- Application consumes 1.5GB memory

From a storage I/O perspective:

• Guest swap partition: Uses the VM's virtual disk (VMDK) for swapping operations
• Hypervisor swapping: Leverages the ESX host's swap file (typically on VMFS datastore)

Performance metrics show:

# Sample performance comparison (lower is better)
+---------------------+---------------+---------------+
| Metric              | Guest Swap    | Hypervisor    |
+---------------------+---------------+---------------+
| Swap latency (ms)   | 12.4          | 8.7           |
| Throughput (MB/s)   | 87            | 120           |
| CPU overhead (%)    | 15            | 9             |
+---------------------+---------------+---------------+

The Linux kernel's swappiness parameter (vm.swappiness) behaves differently:

# Current swappiness value
cat /proc/sys/vm/swappiness

# Recommended settings:
# With guest swap: vm.swappiness = 60 (default)
# Without guest swap: vm.swappiness = 10

Based on real-world testing:

1. For performance-critical VMs: Disable guest swap and increase allocated RAM
2. For memory-constrained environments: Use a minimal swap partition (25% of RAM)
3. For predictable workloads: Configure memory reservations in ESX

Here's how to properly configure a CentOS VM without swap:

# Remove existing swap
swapoff -a
sed -i '/swap/d' /etc/fstab

# Adjust kernel parameters
echo "vm.swappiness = 10" >> /etc/sysctl.conf
echo "vm.overcommit_memory = 1" >> /etc/sysctl.conf
sysctl -p

# Verify settings
grep -i swappiness /proc/sys/vm/

Key ESX metrics to watch:

# ESX CLI commands for memory monitoring
esxtop (then press 'm' for memory view)
esxcli system vm process list --mem-usage
vim-cmd hostsvc/hosthardware | grep memory

When configuring Linux VMs on VMware ESXi, administrators face a fundamental choice: whether to allocate a swap partition within the guest OS or rely solely on VMware's memory management. Both approaches involve swapping, but at different architectural layers:

# Scenario 1: Guest-level swap (traditional approach)
/dev/sda1 / ext4 defaults 0 1
/dev/sda2 swap swap defaults 0 0

# Scenario 2: No guest swap (VMware-managed)
/dev/sda1 / ext4 defaults 0 1

The memory hierarchy operates differently in each case:

Modern Linux kernels (4.0+) implement these memory management behaviors:

• Swappiness: The vm.swappiness parameter (default 60) still affects guest-level swapping even when no swap partition exists
• Balloon driver: VMware Tools includes vmmemctl which coordinates with ESXi host memory management
• Memory compression: ESXi 6.5+ implements memory compression before swapping to disk

# Check current swappiness (works with or without swap partition)
cat /proc/sys/vm/swappiness

# Temporary adjustment (root required)
sysctl vm.swappiness=10

Based on workload characteristics:

1. For memory-intensive workloads:

   Allocate 50-100% RAM as swap (e.g., 8GB RAM → 4-8GB swap)

2. For stateless/containerized workloads:

   Disable swap completely and set swappiness=0

3. For general-purpose servers:

   Minimum 1GB swap or 0.5×RAM (whichever is smaller)

Essential diagnostics for both approaches:

# Guest swap usage
free -h
vmstat 1 5

# ESXi host-level swap
esxtop (then press 'm' for memory view)

# Detailed memory breakdown
cat /proc/meminfo | grep -E 'Swap|Mem'