167 questions from 20 interviews · updated from reports 2016-2025
About
Nutanix is a software company that provides hyperconverged infrastructure, cloud management, and platform tools for running applications across data centers and public clouds. In India, it commonly hires for software engineer, site reliability engineer, and software engineering intern roles.
The roles that come up most are Site Reliability Engineer, Software Engineer and Software Engineering Intern. This covers 20 candidate interviews reported from 2016 to 2025. Most sat it at internship level (10 of 17 that recorded a level). Among the 15 that recorded either route, arrivals split between campus drives (12, 80%) and off-campus applications (3, 20%). Most questions fall under CS fundamentals and DSA.
Ans. An HDD stores data on spinning magnetic disks, while an SSD stores data in flash memory with no moving parts. SSDs are much faster for booting, loading files, and random access, and are more shock resistant. HDDs are usually cheaper per gigabyte and better for large low-cost storage.
Ans. Sort the intervals by start time, then scan once, keeping a result list of merged intervals. For each interval, compare its start with the end of the last interval in the result. If they overlap, extend the end; otherwise, append it. Time complexity is O(n log n) due to sorting, with O(n) space.
Ans. System booting is the process of starting a computer, initialising hardware, loading the operating system kernel, and handing control to it. When powered on, firmware such as BIOS or UEFI runs POST, finds a bootable device, loads a bootloader, which then loads the kernel into memory and starts system services.
Ans. I would write a test class with isolated, repeatable tests covering the normal case, edge cases, invalid input and expected failures. Each test should set up known data, call one behaviour, then assert the result and any side effects. Use simple fixtures or mocks for dependencies, and keep tests independent so order does not matter.
Ans. NTP is a protocol used to synchronise the clocks of computers over a network. It lets machines agree on accurate time by querying time servers, usually in a hierarchy. The key detail is that it estimates network delay and clock offset, so systems can adjust their local clocks reliably.
Ans. JVM OutOfMemoryError occurs when the JVM cannot allocate required memory after trying to reclaim space, usually because the heap is full. It can also happen in metaspace, direct memory, or native memory, or when too many threads are created. The key point is that memory limits are exhausted, not just normal GC activity.
Ans. Identify the question type first, then write down the given facts clearly. Convert words into equations, tables, diagrams, or sequences where useful. Eliminate impossible options and check units, order, and conditions carefully. For reasoning puzzles, test one assumption at a time and verify the final answer against every statement.
Ans. Use a min heap containing the current smallest unmerged element from each array. Insert the first element of every non-empty array with its array index and position, repeatedly pop the minimum, append it to the result, then push the next element from that same array. Time is O(N log k), space is O(k).
Ans. Common OSI protocols include Ethernet at the data link layer, IP at the network layer, TCP and UDP at the transport layer, and HTTP, HTTPS, DNS, SMTP and FTP at the application layer. The key detail is that each layer provides services to the layer above while hiding its own implementation details.
Ans. A router is a networking device that forwards data packets between different networks, choosing the best next hop using routing information. For example, when a laptop at home opens a website, the home router sends the request from the local Wi-Fi network to the ISP network and returns the website’s response to the correct device.
Ans. Run DFS over all vertices and keep the vertex that starts the last DFS; this is the only possible mother vertex. Then run one more DFS from that candidate and check whether every vertex is reached. Use an adjacency list and a visited array. The time complexity is O(V + E).
Ans. Design it as services for menu, cart, ordering, payment, kitchen workflow, delivery dispatch, tracking and notifications behind an API gateway. Store orders in a relational database, cache menus, use queues for order events, and make payment/order creation idempotent. The key detail is reliable asynchronous state changes from placed to delivered.
Ans. chmod changes file and directory permissions in Linux. Permissions control read, write, and execute access for user, group, and others. It supports symbolic mode, such as chmod u+x file, and numeric mode, such as chmod 755 file. Common variations include recursive changes with -R and special bits like setuid, setgid, and sticky bit.
Ans. A garbage collector de-allocates memory by finding objects that are no longer reachable from live program roots, such as stack variables, registers and globals, then reclaiming their space. The key detail is reachability, not whether an object is “unused” logically. Many collectors mark reachable objects, then sweep or compact the rest.
Ans. First compare the file sizes, then compare their contents byte by byte or by computing a strong hash for each file. If sizes differ, the files are not the same. If hashes match, they are almost certainly identical, but a byte-by-byte comparison is the definitive check because hashes can theoretically collide.
Ans. A process is commonly in one of five states: new, ready, running, waiting or blocked, and terminated. A PCB, or Process Control Block, is the operating system’s record for a process. It stores the process ID, current state, program counter, CPU registers, scheduling data, memory details and open file information.
Ans. I would test the main user journeys through the full system, from UI or API entry point to data persistence and external integrations. I would start with critical paths, use production-like data and environments, automate stable flows in CI, and keep a few exploratory checks for edge cases, failures, and usability.
Ans. HTTP sends data in plain text, while HTTPS is HTTP over TLS, so data is encrypted, authenticated and protected from tampering. HTTPS works by using a TLS handshake where the server presents a certificate, the client verifies it, and both sides agree session keys for encrypted communication.
Ans. Hyperconvergence combines compute, storage, networking and virtualisation into a single software-defined platform running on standard servers. Nutanix uses it by clustering nodes and pooling their local disks into shared distributed storage, managed through one control plane, so organisations can scale by adding nodes instead of buying separate SAN and server infrastructure.
Ans. Common Linux commands include ls, list directory contents; cd, change directory; pwd, print working directory; cp, copy files; mv, move or rename files; rm, remove files; mkdir, make directory; cat, concatenate and display files; grep, global regular expression print; chmod, change file mode. They are mainly used for navigation, file handling and permissions.
Ans. Use an explicit stack to simulate the recursive inorder traversal: keep moving to the leftmost node, pushing each node, then pop, visit it, and move to its right child. The stack holds ancestors whose left side is done or pending. Time complexity is O(n), and space complexity is O(h).
Ans. You cannot uniquely construct a binary tree from only its inorder traversal. Many different trees can have the same inorder order. To build a unique tree, you also need another traversal, usually preorder or postorder, or extra constraints such as it being a BST. With inorder plus preorder, use a value-to-index map and recurse in linear time.
Ans. ARP maps an IP address to a MAC address on a local network, while TCP provides reliable, ordered delivery of data between applications. ARP works at the link/network boundary and is local to a LAN. TCP works at the transport layer, using connections, acknowledgements, retransmission, and flow control.
Ans. I would debug ping by checking whether the problem is name resolution, local networking, routing, firewall rules, or the remote host. First ping an IP address such as the default gateway, then an external IP, then a hostname. The key detail is separating DNS failure from actual network reachability failure.
Ans. Architecture should start from clear requirements, scale targets, availability needs, data model, and main user flows. The most important detail is choosing boundaries that keep components independently scalable and maintainable, such as API services, storage, queues, caches, and observability. Also cover failure handling, consistency, security, deployment, monitoring, and cost trade-offs.
Ans. Cloud computing is the delivery of computing resources such as servers, storage, databases, networking and software over the internet on demand. The key idea is elastic, pay-as-you-use infrastructure managed by a provider. Common follow-ups include IaaS versus PaaS versus SaaS, public versus private cloud, scalability, availability, security and vendor lock-in.
Ans. The OSI layers map application communication from user features down to physical transfer: applications use layer 7 for HTTP, SMTP or DNS, layer 6 for encryption and encoding, layer 5 for sessions, layer 4 for TCP or UDP, layer 3 for routing, layer 2 for local delivery, and layer 1 for signals.
Ans. Subnetting divides a larger IP network into smaller logical networks by using a subnet mask or CIDR prefix to separate the network and host parts of an address. IP addressing includes IPv4 and IPv6, public and private addresses, and static or dynamic assignment. CIDR is the modern approach, replacing fixed class-based addressing.
Ans. Serialize a binary tree with a preorder or level order traversal including null markers, then deserialize by reading the stream in the same order and rebuilding nodes recursively or with a queue. For a BST, preorder alone can be enough if deserialisation uses min and max bounds. Both take O(n) time and O(n) space.
Ans. The OSI model is a seven-layer framework for network communication: physical, data link, network, transport, session, presentation and application. Examples are Ethernet physical signalling, Ethernet or PPP at data link, IP and ICMP at network, TCP and UDP at transport, RPC or NetBIOS at session, TLS or JPEG at presentation, and HTTP, DNS, SMTP or FTP at application.
Ans. Choose the smallest subnet mask that gives enough host addresses, where usable hosts are 2 to the host bits minus 2. For example, 50 hosts need 6 host bits because 64 minus 2 is 62, so the prefix is /26. Then increment subnets by the block size in the relevant octet.
Ans. Static IP-to-hostname mappings are stored in human-readable form in /etc/hosts. Each entry has an IP address followed by one or more hostnames. Linux name resolution usually consults this file according to /etc/nsswitch.conf, often before querying DNS, while active network addresses are held by the kernel and shown with ip addr.
Ans. The OSI model is a seven layer reference model for how network communication is structured: physical, data link, network, transport, session, presentation and application. Physical sends bits, data link handles frames and MAC addressing, network routes packets, transport provides end-to-end delivery, session manages conversations, presentation formats or encrypts data, and application serves user protocols.
Ans. Use top or ps to check process performance on a UNIX system. top gives a live view of CPU use, memory use, process state, priority and load average, while ps shows a snapshot for specific processes. For deeper system context, use vmstat, iostat or sar to check CPU, memory, disk and I/O trends.
Ans. Use a regular expression that matches one or more valid local characters, an at sign, a domain name, a dot, and a valid top-level domain. Scan the string with the regex and return the first match or all matches. Store results in a list. Time complexity is linear in the input length for a practical regex.
Ans. Cloud computing means using remote compute, storage, networking and managed services on demand, usually with scalability, availability and pay as you go pricing. If a system crashes, first check monitoring alerts, logs, recent deployments, resource usage and dependencies. The most important detail is to isolate whether the fault is application, infrastructure or external service related.
Ans. The OSI model is a seven-layer framework for describing network communication: Physical, Data Link, Network, Transport, Session, Presentation, and Application. Physical sends bits over media. Data Link handles frames and MAC addressing. Network routes packets using IP. Transport provides end-to-end delivery. Session manages conversations. Presentation formats and encrypts data. Application supports user-facing protocols.
Ans. Windows is more GUI focused and uses Command Prompt or PowerShell, while Linux is more command-line centred and uses shells like Bash. Linux is often preferred for servers, development, automation and scripting because its tools are consistent and powerful. Windows is commonly chosen for desktop use, enterprise software, gaming and Microsoft ecosystem support.
Ans. Cloud computing provides compute, storage, networking and managed services on demand, usually with elastic scaling, pay as you go pricing and shared responsibility for security. In cloud environments, systems must expect failure, latency and changing capacity, so they should be stateless where possible, observable, automated, resilient, horizontally scalable and designed for recovery.
Ans. Common CPU scheduling algorithms include FCFS for simple batch systems, SJF or shortest remaining time for minimising average waiting time, priority scheduling for importance-based work, round robin for interactive time-sharing, and multilevel queues for mixed workloads. The key trade-off is fairness versus response time, throughput, starvation risk, and context-switch overhead.
Ans. The OSI model is a seven-layer way to describe how data moves from one computer to another over a network. Each layer has a job, from physical signals and cables up to applications like browsers. The key idea is separation: each layer relies on the one below and serves the one above.
Ans. No, we do not implement custom memory garbage collection in the current system. We rely on the language runtime for heap management, but we do have scheduled cleanup jobs for expired sessions, old audit records, and temporary files, using retention rules and indexed timestamps to keep deletion efficient and predictable.
Ans. The operating system detects the device, identifies it, loads a suitable driver, and makes it available to applications. Typically the bus controller raises an interrupt or event, the OS enumerates the device, reads its identifiers, matches them to a driver, assigns resources such as addresses or IRQs, and registers the device in the system.
Ans. A public cloud is shared cloud infrastructure run by providers like AWS or Azure, while a private cloud is cloud infrastructure dedicated to one organisation. A data centre is the physical facility with servers, storage, networking, power and cooling. A hypervisor lets multiple virtual machines share one physical server safely and efficiently.
Ans. A network switch connects devices within a local network and forwards data only to the device that should receive it. It learns MAC addresses and maps them to ports. For example, in an office, when one PC sends a file to a printer, the switch sends that traffic only to the printer port, reducing unnecessary traffic.
Ans. I would not send data directly to the broken system; I would write it to durable storage or a message queue and process it later. The key detail is reliability: persist each item, acknowledge only after successful processing, retry with backoff, and move repeatedly failing items to a dead-letter queue for investigation.
Ans. Data transmission happens by converting data into signals, sending them over a physical or wireless medium, and reconstructing them at the receiver. The data is split into packets, each with addressing and control information. Signalling defines how bits are represented, such as voltage changes, light pulses, or radio waves, and synchronised between devices.
Ans. A hard link is another directory entry pointing to the same inode, while a symbolic link is a separate inode whose contents store a path to another file. Hard links share metadata and file data, so either name works if the other is removed. A symlink can break if its target path disappears.
Ans. I would first verify the power path: wall socket, power cable, adaptor or PSU switch, then try a known working outlet and cable. The key detail is to isolate whether power reaches the motherboard. If not, suspect the PSU or adaptor. If it does, check the power button, front-panel connector, RAM seating, and motherboard faults.
Ans. Switches connect devices within the same local network, while routers connect different networks and direct traffic between them, such as between a home network and the internet. A switch usually uses MAC addresses to forward frames locally. A router uses IP addresses, makes routing decisions, and often provides NAT, firewalling, or DHCP.
Ans. Ping checks whether a host is reachable and how long packets take to return, while traceroute shows the path packets take through routers to reach it. For example, if a website is slow, ping can show latency or packet loss, and traceroute can reveal where delays or failures occur, such as at an ISP hop.
Ans. Yes, a guest can be configured with more virtual memory than the host has physical RAM, using memory overcommit. The hypervisor maps guest memory to host RAM only as needed and can reclaim pages, swap to disk, or use ballooning. If the guest actively uses too much, performance drops sharply or allocation fails.
Ans. A tree rotation changes parent-child links while preserving the in-order sequence. For a right rotation at node y, let x be y’s left child, move x’s right subtree to y’s left, then make y the right child of x. A left rotation is symmetric. Each rotation is local and takes O(1) time.
Ans. Use a circular buffer protected by one mutex, plus two counting semaphores: empty initialised to buffer capacity and full initialised to zero. A producer waits on empty, locks, inserts, unlocks, then signals full. A consumer waits on full, locks, removes, unlocks, then signals empty. Each operation is O(1).
Ans. Mount it by installing the NFS client, creating a mount point, then running mount -t nfs server:/export /mnt, or adding the same entry to /etc/fstab for persistence. If it fails, check DNS or ping, showmount -e server, export permissions, client IP access, firewall port 2049, NFS service status, and dmesg or journal logs.
Ans. There is no finite maximum unless cloth thickness or a minimum foldable size is specified. In the usual idealised model, each fold halves one dimension, so you can keep folding after it fits. If the intended question is minimum folds, try both orientations and take min of ceil(log2(n/p)) plus ceil(log2(m/q)), with negatives treated as zero.
Ans. Use POST /pastes to create text, GET /pastes/{key} to read, DELETE for owners, storing key, content, user_id, expiry, visibility and timestamps. Use DynamoDB or Cassandra for scale and TTL, but weaker ad hoc querying; SQL is simpler but harder to shard. Cache hot pastes in Redis. LRU uses hashmap plus doubly linked list, O(1).
Ans. Pick a simple, practical example showing a structured approach: confirm the problem, check power and water, inspect obvious faults, isolate variables, consult the manual, and escalate if needed. Emphasise safety, calm prioritisation, and clear communication. Interviewers listen for logical thinking, not coffee expertise, plus knowing when to ask for help.
Ans. Pick a real example where you followed a logical process: check power, paper, ink or toner, cables or Wi-Fi, error messages, queue status, drivers, and network access. Emphasise calm communication, isolating the fault, documenting steps, and escalating when needed. Interviewers listen for structure, user focus, and not guessing randomly.
Ans. Pick a real situation with pressure, conflict, or uncertainty, where teamwork affected the outcome. Emphasise how you clarified priorities, listened to concerns, assigned ownership, kept communication calm, and adapted when needed. Interviewers listen for self-awareness, accountability, emotional control, practical collaboration, and evidence that you helped the team perform better.
No question matches those filters. Reset filters.
Showing 60 of 167 questions. Ranked by how often the same question came back across interviews.
A spoken interview built from these questions, scored when you finish; the feedback is yours.
Start practisingWhen you are ready, record The One: a single interview hiring teams watch, so you stop repeating first rounds.
Candidate interviews most often cover CS fundamentals (70%) and DSA (25%).
Candidate interviews show an average of 3.9 rounds per experience, with a typical sequence of Online test → Technical → Managerial → HR. Individual interview paths can vary.
Among questions with a recorded difficulty, the mix is easy 30%, medium 63%, hard 7%.
