iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Move the workload as a controlled provider exit, not as a simple infrastructure redeploy. First define recovery-point (RPO) and recovery-time (RTO) objectives, inventory every dependency, and build a compatible destination. Replicate or back up data independently of compute and Kubernetes, perform a final quiesced synchronization, validate representative training and inference jobs, switch traffic, and retain the original environment for rollback.
The most important boundary is between infrastructure and state. Azure Resource Manager documentation puts it plainly: “Infrastructure as code (IaC) doesn’t relocate data.” Terraform, ARM templates, or another deployment tool can recreate clusters and networks, but they do not move model checkpoints, object versions, database rows, persistent volumes, secrets, or queue state unless you arrange separate data transfers.
Start by defining what “without losing data” means
A failing neocloud can have several different failure modes: GPU capacity may be unavailable while storage still works; the control plane may be degraded; networking or billing may prevent access; or the provider itself may be at risk of disappearing. Record the exact condition, what remains reachable, and the timestamp of the last known-good workload state. That determines whether you can perform a measured migration or need an emergency recovery.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Set separate RPO and RTO targets for each stateful component. RPO is the maximum acceptable amount of recent work to lose; RTO is how long the service may be unavailable. A training checkpoint, a feature store, an inference database, and an event queue rarely deserve identical targets.
#1 Best Overall
- Powerful Turbo Fan:CAGIWIRU Electric Air Duster have reached speed up to 130000RPM,Can efficient remove dust and debris.With LED light and Three Adjustable Speed to meet with different cleaning tasks.
- With Attached USB C Fasting Charging cable.Charing 3 hours,Enjoy up to 240 minutes of use on the lowest setting, For third speed can use for 25mins.with four charging options to meet with your needs.
- Wide Application: attached 5 different nozzles,making it suitable for a variety of scenes, such as car,pet, pc, keyboards, and other electronic devices. It also serves for office and home clean duster.
- Handy Design & Portable:Cordless air duster electric nozzle is Light weight and small size, design and comfortable to hold and space saving, convenient to carry around. You can put it in your room or inside the car without taking up space.
- Economical &Reusable: Built durable materials with a long-lasting battery,our CAGIWIRU air duster it's a sustainable alternative to disposable air cans.
| Workload state | Questions to answer | Evidence to record |
|---|---|---|
| Training checkpoints | How much training can be repeated? Which checkpoint is the latest usable one? | Checkpoint timestamp, object version, model hash, optimizer and scheduler state |
| Datasets and feature stores | Can the source be reconstructed, or must every object and feature be copied? | Object counts, byte totals, schema versions, partition lists and checksums |
| Inference data | Which requests, embeddings or user records must survive a cutover? | Database high-water mark, queue offsets, transaction logs and retention rules |
| Platform configuration | What must be rebuilt exactly for jobs to schedule and authenticate? | Manifests, image digests, secrets references, policies and endpoint configuration |
Write the RPO and RTO beside each item. A migration is not successful merely because pods start on the new provider; it is successful when the required state is present, consistent and usable within those limits.
Choose cold, warm or hot relocation
Choose the relocation mode according to downtime tolerance and the replication features available for your databases, filesystems and object stores. “Hot” is not a universal option: it requires service-level replication and synchronous data replication where the service supports them.
| Mode | How it works | Best fit | Main trade-off |
|---|---|---|---|
| Cold | Back up data, stop the source, deploy the destination, restore, then validate. | Noncritical jobs, reproducible training, or a planned outage. | Longest outage and data loss up to the backup point. |
| Warm | Keep the source running while asynchronously replicating data and building the destination. Copy the final delta, switch endpoints, validate, then retire the source. | Most production AI services where brief downtime is acceptable. | Replication lag and write ordering must be measured and controlled. |
| Hot | Use supported synchronous replication, activate and validate the destination, then stop the source. | Strict continuity requirements and services that offer synchronous replication. | Higher complexity and cost; impossible when the underlying service lacks synchronous replication. |
For asynchronous disk replication, Google guidance recommends scheduled downtime and taking the workload offline before stopping replication so that in-flight writes do not create data loss. Treat that as a design requirement, not an optional refinement.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Inventory before copying anything
Create a dependency register that another engineer can use during an outage. Microsoft’s migration guidance calls for visibility into architecture, performance, security, code, databases and dependencies; an AI exit plan needs those categories plus GPU and scheduler details.
Compute and accelerator inventory
- Record every machine or node pool, GPU model, GPU count, memory, local scratch disks, region and availability zone.
- Capture driver, CUDA or other runtime compatibility, kernel constraints, node labels, taints, topology rules and autoscaling limits.
- List training, fine-tuning, batch inference, online inference and data-ingestion jobs, including their resource requests and queues.
Kubernetes and platform inventory
- Export cluster versions, namespaces, deployments, stateful sets, jobs, cron jobs, daemon sets, services, ingress objects and persistent-volume claims.
- Include custom resource definitions, operators, Helm charts and values, admission policies, network policies, resource quotas and scheduler configuration.
- Map service accounts, RBAC bindings, workload-identity settings, image registries, image digests, certificates, DNS records, queues and external APIs.
Useful read-only snapshots include kubectl get all -A -o yaml, kubectl get pvc -A -o yaml, kubectl get crd -o yaml and a Helm release export. Treat any output containing secret data as sensitive: store encrypted copies, redact routine tickets and never commit plaintext credentials to a repository.
Data and control-plane inventory
- Identify object stores, block volumes, shared filesystems, databases, vector or feature stores, caches and message queues.
- For each store, record size, object or row counts, versioning, lifecycle rules, retention or legal holds, encryption keys, ACLs and replication status.
- List identity providers, federation settings, service principals, certificates, firewall rules, private routes, load balancers, API gateways, observability agents and alert destinations.
If the source is a nonstandard environment, Azure Migrate documents agent-based replication for machines running in AWS, Google Cloud, Xen, KVM and private clouds when agentless prerequisites are unavailable. Select an agent only after checking its network path, permissions and operating-system support.
Build and qualify the destination before the outage
Provision the target account, network and security boundaries before copying production state. Use the inventory to create a compatibility checklist rather than assuming that a similarly named GPU or Kubernetes service behaves the same way.
Rank #2
- Screen Cleaner is Your Screen’s New Bestfriend! - EVEO is proud to present a screen cleaner that’s perfectly suited for your TV. Meet your new screen cleaner, computer screen cleaner, laptop cleaner, iPad cleaner, Macbook,tv cleaner or any other electronic device. Our screen cleaner spray and wipe along with the included plush microfiber cloth, easily removes dust, fingerprints & other smudges on the screen’s surface. Clean effectively for a streak-free screen
- Gentle on your Screens - Our screen cleaner spray and wipes is compatible to be used even for the most sensitive LCD, LED, CRT, and OLED screens in proper use. Be at ease knowing that this product is compatible with all the major brands: LG, SONY, Samsung, Sharp, iPad, iPhone, Android screens, Kindle, PC monitors - we’ve got you covered! Can be used as LCD screen cleaner, tv cleaner, smart tv screen cleaner, and more
- Premium Super Soft Plush Microfiber Cloth - Included in this efficient screen cleaning kit. This screen cleaner spray and wipe has our signature plush microfiber cloth specially designed to comply with screens and monitors and leave them streak-free. Has been shown to effectively absorbs stubborn stains, zero streaks, and no fiber shedding. Always use this together with our screen cleaner for the best results
- Easy-to-Use Screen Cleaner Kit + microfiber cleaning cloth - The EVEO Screen Cleaning Kit will easily be your new favorite tool when it comes to taking care of the things you love most, quick and easy solution for a streak-free screen. Simply screw on the sprayer gun on the screen cleaner bottle. Spray the cleaning liquid onto the microfiber cloth, then wipe the screen until desired cleanliness & shine is achieved.
- Quality Screen Care Kit Worth Purchasing - Enjoy a convenient Screen Cleaner anytime, anywhere you need it. Your satisfaction is our top priority. We stand behind the quality of our products and that’s why, you shouldn’t be worrying at all. We manufacture the Best Screen Cleaner. and can be used for laptop cleaning kit, laptop screen cleaner, tv screen cleaner for smart tv. MacBook screen cleaner, monitor cleaner or screen cleaner spray for electronic devices with microfiber cleaning cloth
| Area | Acceptance test |
|---|---|
| GPU platform | The required GPU model or an explicitly qualified substitute is schedulable; drivers and runtimes load; a smoke test sees the expected device count and memory. |
| Kubernetes | Required cluster version, CRDs, operators, admission controls, storage classes and network policies apply without errors. |
| Storage and databases | Target throughput, latency, capacity, snapshots, encryption and recovery features meet the workload’s RPO and RTO. |
| Identity | Service accounts obtain only the intended permissions, workload identity works, and secret references resolve without exposing values. |
| Networking | Private routes, egress controls, DNS, certificates, queues and external APIs work from the new nodes. |
| Operations | Metrics, logs, traces, alerts, dashboards, on-call contacts and audit records are visible from the destination. |
Prefer pinned image digests and reproducible infrastructure definitions. NVIDIA’s AI-cloud requirements emphasize explicit node-type selection, default labels and taints, standard Kubernetes service accounts with workload identity, Terraform where appropriate, and backup and disaster-recovery planning. These controls prevent a migration from silently changing scheduling, identity or recovery behavior.
Move data independently of infrastructure
Provision destination databases, object stores and volumes first, then transfer state through a mechanism appropriate to each data type. Keep the source available until the final validation and rollback window has expired.
Databases and transactional stores
Use a provider-supported migration or replication feature when one exists. Establish the target schema and extensions, copy a consistent baseline, stream changes, and record the source transaction or log position. Before cutover, stop or quiesce writers, wait for the target to reach that position, and verify row counts, key ranges, indexes and application-level invariants. A row count alone cannot prove that the same records arrived in the same state.
Object storage and checkpoints
Copy model checkpoints, tokenizer files, datasets, feature snapshots and configuration artifacts with their object versions and metadata. Google’s Storage Transfer Service, Transfer Appliance and cross-bucket replication are examples of mechanisms for this class of transfer. Cross-bucket replication copies new and updated objects asynchronously between independent buckets, so monitor lag and define how deletes are handled.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteFor every important prefix, record object count, total bytes and a manifest containing path, size, version or generation and checksum. Recalculate checksums at the destination. For checkpoints, also open the file with the training framework, load the model and optimizer state, and compare a cryptographic model hash where the format permits it.
Filesystems and persistent volumes
Choose a snapshot, volume-replication or file-transfer method that preserves permissions, timestamps, sparse files, symlinks and application-consistent state. Do not copy a live database directory with a generic file utility unless the database documentation explicitly supports that procedure. For large volumes, measure transfer throughput and remaining replication lag rather than relying on a nominal bandwidth figure.
Queues, caches and ephemeral data
Decide explicitly whether each queue, cache and temporary artifact is durable state or disposable state. Export consumer offsets or replay from a defined event position when ordering matters. Rebuild caches after cutover instead of treating stale cache contents as authoritative.
Rank #3
- 【Large size】：Measuring 800 x 300 mm/ 31. 5×12 inches, which is wider and taller than others. Large mouse pad can perfectly fit your keyboard and mouse of any size, and there is also ample space for peripherals such as phones, tablets, etc.
- 【LED backlight mouse pad】：Blade Hawks RGB mouse pad provides 7 static modes and 7 dynamic modes will give you the options you are looking for. Power-off memory function remembers the last lighting mode you selected. The super glow fiber Chroma will give you a colorful gaming setup that you want.
- 【Anti-slip rubber base】：Made from Natural rubber, this gaming mousepad will firmly grip your desktop, allowing you to enjoy the ultimate precision in the games you are most passionate about. It also comes with a texture that enhances this ability even further.
- 【Smooth and waterproof surface】: Micro-textured cloth surface, for extremely precise mouse control and a smooth movement. When liquid splashes on the gaming mouse pad, it forms water droplets and slides down. Will not delay your work or gameplay.
- 【Plug and play】: This mouse and keyboard pad is powered by USB, plug and play, no driver required. One button control light modes. Press one time to change the lighting mode, press twice to change the brightness, press and hold about 3 seconds turn ON/OFF.
Replication is not a complete backup
Replication reduces the risk of an infrastructure failure, but it does not protect against corruption, user error or malicious attacks. Google Compute Engine documentation states: “Replication protects against infrastructure failure by maintaining a continuously updated copy of your data in another location. However, replication doesn’t protect against data corruption, user error, or malicious attacks.”
Recommended Free Tools
Keep an isolated or access-separated backup in addition to live replication. Use immutable retention where possible, protect the backup account with separate credentials, and rehearse a restore to an environment that is not the production target. A restore rehearsal should prove that keys are available, permissions are correct, metadata survives and an AI job can actually read the restored checkpoint.
Match the protection method to the failure domain. Synchronous regional disk replication can provide zero-RPO protection for a zonal failure when supported. Asynchronous cross-region replication is more suitable for regional disaster recovery, but it requires a defined lag budget, consistency groups and a tested failover procedure. Neither method removes the need for point-in-time backups.
Preserve the AI and Kubernetes state that is easy to overlook
Scheduling and accelerator compatibility
Recreate GPU node labels, taints, tolerations, topology constraints, device-plugin configuration, autoscaling rules and queue priorities. A workload can be “running” while silently using a different GPU class, fewer devices or an incompatible driver. Run a representative training step and an inference benchmark that exercises the same kernels, batch sizes and precision modes used in production. Record expected outputs and failure thresholds before the test.
Images, models and software supply chain
Copy images to a registry reachable from the destination and deploy by immutable digest rather than a floating tag. Verify CUDA or other accelerator runtime compatibility, Python and system-library versions, tokenizer files, model configuration and custom operators. Keep a bill of materials or lock file so that a rebuild after rollback uses the same bits.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesIdentity, secrets and certificates
Recreate service accounts, RBAC, workload identity, external-secret references, certificate chains and key access. Do not assume that an identically named account in the new provider has the same subject, permissions or audit behavior. Rotate credentials that were exposed during the incident, but do so in an order that keeps both source and destination usable during the transition.
Persistent-volume and edge-device data
Export persistent-volume definitions and storage-class parameters, then test attach, mount, read and write operations on the destination. For edge or GPU devices, separate cloud-tiered shares, local shares and virtual-machine data. Azure warns that local data not tiered off-device may be lost after a non-tolerable failure; copy any required local state before shutting down or losing access to the device.
Rank #4
- 【Softer and More Comfortable】Vaydeer wrist rest has unique diamond pattern, which is the combination of softness and aesthetics. The materials of wrist rest are improved into higher quality memory foam and covered with silky smooth lycra. The computer wrist rest makes you as comfortable and cushiony as like rest your wrists on clouds.
- 【Ergonomic Wrist Saver】The wrist rests for keyboard and mouse comes with a 17.32×3.15×0.83 inch keyboard wrist pad and a 5.94×3.15×0.83 inch mouse wrist support. Based on ergonomic design, the unique concave shape is the perfect fit for your wrist joints. The wrist rest pad fits most computer keyboards and laptops, improve hand and wrist posture, release your wrist and arm stress.
- 【Non-Slip Rubber Bottom】Featuring an anti-skid silicone base on the bottom, this wrist keyboard support stays firmly in place on your desk, preventing the padding from sliding around, ensuring stable and consistent wrist support during extended computer sessions.
- 【Better Experience & Pain Relief】Our keyboard arm rest is beneficial to alleviate the soreness caused by direct contact and friction between your arm and a hard desk surface, reducing the risk of wrist fatigue or carpal tunnel. The soft texture of memory foam can evenly distribute the pressure around your wrists and provide good support with just enough give.
- 【Helpful in Multiple Scenarios】Whether you're working, studying, writing, typing, gaming, this keyboard and mouse rest combo is an essential accessory to add comfort and support to your hands and wrists. It’s also a great gift for men, women, family, friend, coworker, gamer, teacher, etc.
Observability and operational controls
Move dashboards, alert rules, log parsers, trace sampling, quotas and audit sinks. Confirm that alerts identify the new cluster and that on-call staff can distinguish a migration-induced error from a model or data error. An unmonitored destination is not ready for production even if its jobs pass a smoke test.
Test the destination with production-shaped work
Run tests in increasing order of risk:
- Platform test: schedule a pod on every required GPU type, mount each storage class, resolve a secret reference and reach required endpoints.
- Data test: compare manifests, checksums, object versions, database counts, schema checks and queue offsets against the source snapshot.
- Checkpoint test: restore the latest usable checkpoint, run one or more training steps, save a new checkpoint and reload it.
- Inference test: send representative requests, compare model outputs within an agreed tolerance, and measure latency, throughput and error rates.
- Recovery test: delete a disposable target workload, restore it from the documented backup, and time the operation against the RTO.
Document pass or fail results, not just screenshots. A failed test should block cutover until its impact is understood or an explicit risk owner accepts it.
Free tools Windows power users keep installed
One-click scans. No signup required.
Execute the final cutover
Use a written change plan with named owners, timestamps and stop conditions. The sequence below limits writes that could otherwise outrun replication.
- Announce the window. Notify users and operators, freeze nonessential deployments and stop jobs that create untracked state.
- Quiesce writers. Pause training checkpoint uploads, feature-store updates, database writes, file synchronization and queue producers as required by the consistency design.
- Wait for the lag threshold. Confirm that every replication stream is at or below the agreed RPO and record the final source timestamp, transaction position and object-generation watermark.
- Create and protect the final checkpoint. Flush application buffers, complete a final model checkpoint, snapshot or transaction boundary, and verify that it is durable at the destination.
- Stop replication only after durability is proven. Preserve logs and manifests showing that the final delta arrived. For asynchronous disks, take the workload offline before stopping replication when the provider procedure requires it.
- Switch traffic. Update DNS, load balancers, API gateways, queue consumers and client configuration in the planned order. Reduce DNS TTL in advance where practical, but allow for resolver caching.
- Validate immediately. Check record counts, checksums, model hashes, checkpoint readability, permissions, queue behavior, latency, error rates and representative training and inference jobs.
- Keep the source recoverable. Isolate it from new writes, preserve its snapshots and credentials under controlled access, and do not destroy it until the rollback criteria and retention period expire.
Define rollback before switching traffic
Rollback is safe only while the source remains a coherent copy. Decide in advance which symptoms trigger reversal: missing objects, checkpoint load failure, authentication errors, unacceptable latency, data-integrity mismatches or error rates above the agreed threshold. Set a deadline after which rollback would require a new forward migration because the destination has accepted new writes.
If rollback is triggered, stop destination writers, capture any newly created state, restore the source endpoint, and reconcile or preserve destination-only records according to the application’s data policy. Do not point traffic back to a source that continued receiving untracked writes during the cutover.
Make the exit plan permanent
A provider exit plan should be maintained while the service is healthy, not written during the outage. Document:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Export APIs, file formats, database dump formats, object metadata and Kubernetes manifests.
- Who owns encryption keys, how keys are escrowed, and how a restore works if the provider account is unavailable.
- Identity federation, service-account recreation, certificate renewal, private routes, firewall rules, DNS and queue endpoints.
- Data residency, retention, legal holds, deletion procedures, egress limits and transfer-appliance logistics.
- Monitoring, audit logs, support contacts, escalation paths and the exact last-known-good timestamp to use during an incident.
- A schedule for destination testing, restore rehearsals, checksum verification and updating image and driver compatibility records.
Google publishes switching procedures and an exit program covering services such as BigQuery, Cloud Storage, Cloud SQL, Filestore, Spanner and Persistent Disk. Use the equivalent export and switching documentation for every service in your own stack, and test the documented procedure rather than treating its existence as proof of portability.
Best Value
- 【No Drill Mounted】Cable management tray can be assembled in as little as 3 minutes or less-simply. Side clips easily clampe the edge of desk, pass through the holes on the sides, and secure excess cables with baskets. Desk cable management supports inward or outward installation to meet your needs in different scenarios, and provides you with great convenience when collecting and organizing wires.
- 【Rubber Pad Protects Desk】Wire organizer under desk built-in soft rubber pads on the clamp protect your desk from scratches, so you will not damage your furniture or office.
- 【Sturdy and Beautiful】Cord management under desk is made of high-quality carbon steel and holds about 15 lbs. The wire basket has 2 holes on each side of the tray to enable your cables to pass freely. Cable baskets are suitable for desk thickness use from 0.4" to 2.4".
- 【Space Saving & Keep Organized】Desk cable management tray holds all your power cords, outlet strips, USB hubs and computer transformers hidden from your feet, and the data cables will not enter the vacuum cleaner to keep your desk clean and tidy. Desk wire organizer is also suitable for use in the kitchen and outdoors.
- 【Reduce Safety Risks】Cable tray under desk keeps all plugs away from your kids, no more worry about tripping. Using 2 holes wire tray saves extra space, helps you easily cabling and wire protection. The mesh design is not easy to accumulate dust, but avoids safety accidents caused by messy wires.
How to compare replacement providers
Evaluate candidates against the workload’s actual dependencies instead of comparing GPU hourly rates alone.
| Comparison axis | What to verify |
|---|---|
| GPU capacity | Availability of the required model, memory and interconnect; driver and runtime compatibility; reservation and quota process. |
| Data portability | Database import and replication options, object-storage APIs, volume snapshots, metadata preservation and checksum support. |
| RPO and RTO | Achievable lag, synchronous options, point-in-time restore, consistency groups and documented failover time. |
| Kubernetes and IaC | Supported versions, CRDs, storage classes, workload identity, node labels and taints, and reproducible provisioning. |
| Security and identity | Key ownership, federation, least-privilege roles, private networking, audit logs and data-residency controls. |
| Operations | Metrics, logs, traces, alert integrations, support response and access during a provider incident. |
| Economics and exit | Egress and transfer limits, appliance availability, retention charges, contract terms and the cost of maintaining a second destination. |
| Rollback options | How long the source can remain available, how bidirectional synchronization works, and which writes must be frozen to reverse the cutover. |
Common migration failures and their fixes
“The cluster recreated, so the migration is done.”
Cause: infrastructure definitions were applied without moving state. Fix: run separate database, object, volume, queue and secret-transfer workstreams, each with its own integrity evidence.
“The latest checkpoint exists but will not load.”
Cause: incomplete multipart upload, missing object version, incompatible driver or absent custom operator. Fix: verify the manifest and checksum, restore the entire dependency set, and perform a real checkpoint-restore test before cutover.
“Inference starts, but results or latency changed.”
Cause: a different GPU, precision mode, kernel, batch-size limit or model artifact. Fix: compare node hardware, runtime versions, image digests and model hashes, then run production-shaped requests with recorded tolerances.
“Replication reports healthy, but bad data was copied.”
Cause: replication mirrors corruption, accidental deletion or malicious changes. Fix: retain isolated point-in-time backups, use immutable retention where possible, and rehearse a clean restore.
“Traffic switched, but clients still reach the old service.”
Cause: DNS and client caches, overlooked API gateways or hard-coded endpoints. Fix: inventory every ingress path, lower TTL ahead of the window, monitor both endpoints and keep the source isolated rather than serving conflicting writes.
Quick Recap
Cutover checklist
- Failure mode, last-known-good timestamp, RPO and RTO are written for every stateful component.
- Compute, GPU, Kubernetes, identity, network, storage, database, queue and observability dependencies are inventoried.
- Destination drivers, runtimes, node labels, taints, storage classes, policies and image digests are qualified.
- Data has been copied independently of infrastructure, with checksums, counts, versions, permissions and encryption verified.
- Replication lag, final checkpoint durability and writer-quiescence steps have named owners.
- Representative training, checkpoint restore, inference and recovery tests have passed.
- Traffic-switch steps, validation queries, rollback triggers and source-retention dates are documented.
- An isolated backup and a tested restore remain available after the migration.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
