HCIE-Storage flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

RAID 5 write penalty

Each host write costs four backend I/Os: read old data, read old parity, write new data, write new parity. Random small writes therefore scale poorly versus RAID 10.

RAID 6 write penalty

Each host write costs six backend I/Os because two independent parity blocks must be read and rewritten. It survives two member failures but needs more disks for equal usable capacity.

Usable RAID capacity

Usable capacity equals member count times the smallest member size, minus parity overhead. Mixed-size members are limited by the smallest disk in the group.

Rebuild exposure window

Time to reconstruct a failed member, roughly member capacity divided by rebuild rate. Longer rebuilds raise the chance a second disk fails during recovery.

Thin provisioning ratio

Provisioned logical capacity divided by backing physical capacity. Values above one mean over-subscription, so monitoring and auto-expansion thresholds prevent pool exhaustion.

Copy-on-write snapshot

Before a source block is overwritten, its original content is copied to snapshot space. Space grows with changed blocks, not with snapshot age.

Copy-on-first-write snapshot

The first write to a protected block after snapshot creation is redirected to a new location, and the mapping table records the change. Subsequent writes update in place.

Snapshot versus clone

A snapshot is a point-in-time dependency on its source LUN. A clone is a full independent copy that survives source deletion without promotion.

Write-back cache risk

Acknowledged writes reside in volatile cache until destaged. Without working battery or capacitor protection, arrays switch to write-through to avoid data loss on power failure.

IOPS from queue depth

IOPS equals outstanding I/Os divided by latency in seconds. At one millisecond latency, one outstanding request yields about one thousand IOPS.

ALUA asymmetric access

Target ports report active-optimized or non-optimized states. Hosts send I/O through optimized paths and use non-optimized paths only for failover or explicit preference.

Multipath round-robin

Host distributes I/O across all available paths to a LUN, typically per I/O or per block. It maximizes bandwidth but requires symmetric or well-tuned ALUA support.

Cache hit ratio

Read hits divided by total reads. Higher ratios reduce backend load; sequential streaming workloads usually show low ratios regardless of cache size.

Hot spare versus global spare

A dedicated hot spare serves one disk domain; a global spare can rebuild any failed member in the array. Global spares reduce idle capacity waste.

Wide striping benefit

Spreading a LUN across many disks raises aggregate throughput and IOPS because requests are served in parallel, but it widens the impact of any single disk failure.

LUN expansion methods

A LUN can grow by extending within the same RAID group or by striping across new members. Host file systems must support online growth to use added capacity.

Block zero detection

Arrays detect all-zero writes and skip physical allocation, saving capacity. Thin LUNs benefit most, but the optimization is invisible to hosts.

Storage tiering policy

Data moves between performance and capacity tiers based on access frequency measured over a window. Migration granularity is typically a sub-LUN chunk, not a whole LUN.

SSD wear leveling

Firmware distributes erase cycles across flash blocks to prevent premature failure. Write amplification and garbage collection determine endurance more than raw capacity.

SAS versus NL-SAS

SAS drives spin at 10K or 15K RPM with dual ports for high duty cycles. Nearline SAS runs at 7.2K RPM for capacity, with lower IOPS and higher latency.

FC zoning single-initiator

Each zone contains one host port and its target ports. Single-initiator zoning limits disruption and prevents hosts from discovering or accessing unrelated LUNs.

iSCSI CHAP authentication

Challenge Handshake Authentication Protocol verifies initiator identity with a shared secret. Bidirectional CHAP also authenticates the target to the initiator.

NFS export options

Export controls client access by host, subnet, or netgroup, plus read-write and root-squash settings. Misconfigured no_root_squash grants remote root privileges.

SMB multichannel

Client and server use multiple TCP connections over available NICs for one session. Aggregate throughput rises when links and switch paths are independent.

Object storage consistency

Most object stores provide read-after-write for new objects and eventual consistency for overwrites and listings. Applications must tolerate stale list results.

Erasure coding overhead

With k data fragments and m parity fragments, overhead is m divided by k. A 8+2 scheme uses 25 percent extra capacity and survives two fragment losses.

Synchronous replication RPO

Writes are acknowledged only after the remote site confirms. Recovery point objective is effectively zero, but latency between sites directly limits write throughput.

Asynchronous replication RPO

Writes are acknowledged locally and shipped in cycles. Recovery point objective equals the replication interval plus in-flight data, so it is never zero.

Split-brain in stretched cluster

Both sites believe they are primary after link loss. A quorum witness or tiebreaker site decides which side continues serving to prevent divergent writes.

Consistent snapshot group

Multiple LUNs are snapped at the same logical instant so dependent application data stays mutually consistent. Required for databases spanning data, log, and archive volumes.

Backup 3-2-1 rule

Keep three copies of data, on two different media types, with one copy offsite. It protects against media failure, site loss, and correlated corruption.

Ransomware air gap

An isolated copy is unreachable from production credentials, for example offline tape or an isolated vault. Immutability prevents deletion or encryption by compromised accounts.