When working with ZFS storage pools and file systems, performance degradation at high utilization levels is a well-documented phenomenon. The critical threshold often cited is 80% pool capacity, beyond which:

• Write amplification increases due to COW overhead
• Metadata operations become more expensive
• Resilvering operations slow down significantly

In your specific scenario with a 10T raidz2 pool containing a volume/test filesystem with 5T reservation:

# Sample ZFS creation commands
zpool create tank raidz2 /dev/sda /dev/sdb /dev/sdc
zfs create -o reservation=5T tank/volume/test

The performance impact manifests differently based on which perspective you examine:

From the Parent Filesystem Perspective

When filling volume with ~5T data:

• The pool remains at 50% utilization (5T/10T)
• ZFS can technically use the reserved space for COW operations
• However, some metadata operations may still suffer as they're constrained by the filesystem boundary

From the Child Filesystem Perspective

The reserved space in volume/test behaves differently:

# Checking space utilization
zfs list -o name,used,available,reservation tank/volume
zfs list -o name,used,available,reservation tank/volume/test

Performance characteristics include:

• Write performance remains stable until reservation is nearly exhausted
• Read performance is less affected but may degrade slightly due to fragmentation

In your second scenario with two 3T-reserved filesystems:

zfs create -o reservation=3T tank/volume/test1
zfs create -o reservation=3T tank/volume/test2

Writing 7T to test1 creates an interesting situation:

• The pool is now at 70% utilization (7T/10T)
• test1 exceeds its reservation by 4T
• Performance impact becomes asymmetric:

Filesystem	Performance Impact
test1	Severe degradation (beyond reservation)
test2	Minimal impact (reservation intact)

For optimal performance in reservation-heavy environments:

# Monitoring command examples
zpool iostat -v 5
zfs get all tank/volume | grep -E 'reservation|quota'

Key strategies include:

• Maintain at least 20% free space across the entire pool
• Use reservations judiciously - they're not performance guarantees
• Monitor individual filesystem utilization separately from pool utilization

---

When dealing with ZFS storage allocations, it's crucial to distinguish between pool-level free space and filesystem reservations. The performance impact manifests differently in these scenarios:

# Typical ZFS pool creation with RAIDZ2
zpool create tank raidz2 sda sdb sdc sdd sde sdf
zfs create tank/volume
zfs create -o reservation=5T tank/volume/test

In your first scenario with a 5T reservation for volume/test, writing 5T to the parent volume filesystem:

• Pool-level free space remains 5T (reserved but unused)
• ZFS can utilize the reserved space for COW operations
• Performance degradation would be minimal until actual pool utilization exceeds 80%

The second scenario with two reserved filesystems presents more complex behavior:

zfs create -o reservation=3T tank/volume/test1
zfs create -o reservation=3T tank/volume/test2

Key observations when writing 7T to test1:

• Pool shows 4T free (10T total - 6T reserved)
• Actual used space exceeds one reservation (7T > 3T)
• ZFS dynamically borrows from test2's reservation
• Performance characteristics become filesystem-specific:
  • Write-heavy filesystems show more pronounced slowdown
  • Read performance remains relatively stable

Essential commands to monitor actual performance impact:

# Check pool fragmentation
zpool list -v

# Monitor I/O latency
zpool iostat -lv 1

# ARC and L2ARC statistics
arcstat.py 1

# Filesystem-specific usage
zfs list -o space -r tank/volume

For optimal performance in reservation-heavy environments:

• Maintain at least 20% free space at pool level
• Use refreservation instead of reservation for exact space guarantees
• Consider splitting heavily used filesystems to separate pools
• Monitor zfs_dbuf_cache and zfs_arc_meta_limit parameters

Example of proactive space management:

# Set warning threshold at 75% utilization
zfs set quota=7.5T tank/volume/workload

# Automated alerting script
zfs list -H -o name,used,avail | awk '$3/($2+$3) < 0.25 {print "Warning: "$1" low on space"}'