Glossary
Overview
A reference page that consolidates the key terms and definitions used across the ECI platform.
Compute
Virtual Machine (VM)
An isolated compute environment provided by ECI. You can configure CPU, memory, storage, and network independently. A VM is created by choosing an instance type and an OS image, and goes through a start / stop / restart / delete lifecycle.
Instance Type
A predefined bundle of specs you select when creating a VM. The number of vCores, memory size, and type and count of GPU/NPU devices are all wrapped into a single code name, so instead of picking individual settings, you only need to pick the type that matches your workload. The code-name prefix indicates the family.
| Prefix | Family | Example codes | Description |
|---|---|---|---|
G- | GPU | G-NHHS-640, G-NAHP-80, G-NBTHS-180 | Format: G-{device code}-{total VRAM (GB)}. NHHS is H100 SXM, NAHP is A100 PCIe, NBTHS is B200 SXM |
N- | NPU | N-FR-48, N-FW-16, N-FW-32, N-FW-64 | Format: N-{device code}-{total memory (GB)}. FR is Renegade, FW is Warboy (FuriosaAI) |
C- | CPU | C-2, C-4, C-8, C-16, C-32 | General-purpose CPU. Memory is vCore × 2 GiB |
M- | CPU | M-2, M-4, M-8, M-16 | Memory-optimized CPU. Memory is vCore × 4 GiB |
The portal's Infrastructure > Instance Types menu lists the full set of types currently offered, along with their specs.
Region
A geographically separated infrastructure area. A region is the parent grouping for multiple zones, and each region has its own power and network domain. The portal's Infrastructure > Regions menu shows the regions currently available.
Zone
An independent power, cooling, and network area within a single datacenter. VMs, storage, network, and every other resource belong to exactly one zone, and resources cannot be shared across zones. See Infrastructure > Zones in the portal for the list of zones currently available.
Virtual Cluster
An HPC environment that bundles multiple VMs together over an InfiniBand or RoCEv2 high-speed network fabric. It is used for large-scale distributed training.
Storage
Block Storage
Persistent storage that attaches to a VM as a disk. Both the OS disk and data disks are block storage. It is preserved even after the VM is deleted, and continues to be billed until you delete it.
Object Storage
S3-compatible storage that keeps files in buckets. By default you access it through an API (aws-cli, rclone, etc.), but you can also use rclone mount to mount a bucket as a directory inside a VM and treat it like a regular filesystem. It is well suited for storing large training datasets or sharing data across multiple VMs.
Parallel File System (PFS)
High-performance shared storage that multiple VMs can read and write at the same time. Used for checkpoints and dataset sharing during distributed training.
Snapshot
A point-in-time copy of a block storage volume. Used for recovery or to create a new volume. This is distinct from a block storage image (an OS template).
Snapshot Scheduler
A feature that automatically creates block-storage snapshots on a fixed schedule. You set the schedule with a cron expression and specify a maximum retention count (1 to 100); older snapshots beyond that count are automatically deleted. It is useful for keeping consistent recovery points without having to remember to take manual backups.
Block Storage Image
An OS template used as the starting point for a VM's boot disk. When you create a block storage volume from one of the images Elice provides — such as Ubuntu or a CUDA/PyTorch preinstalled environment — a boot disk is created with that image copied onto it. See Infrastructure > Block Storage Images in the portal for the available list.
Network
Virtual Network
An isolated private network used for VM-to-VM communication. Equivalent to a VPC in AWS; the IP range is defined by a CIDR block.
Subnet
A virtual network split into smaller IP ranges. You specify the gateway IP and CIDR, and VMs receive IPs from a subnet.
Public IP
An external IP that connects a VM to the internet. You need a public IP not only when something outside connects to the VM over SSH or HTTP, but also when the VM needs to reach out to the internet (for example, apt install-style package installs, calls to external APIs, or downloading models and datasets). A public IP is attached to a network interface, not to the VM, and is billed separately from allocation until deletion.
Network Interface
A virtual network card that connects a VM to a subnet. You either create one as part of the network step when creating a VM, or attach an existing network interface. Public IPs attach to the network interface, not to the VM.
Firewall
Security rules that allow or block traffic going in and out of a virtual network. Rules apply to both inbound traffic entering the VM and outbound traffic leaving the VM, and define which communication to allow or deny based on IP, port, and protocol.
The default behavior is allow-all. If no rules are configured, all traffic passes through, so when you need security you typically add ACCEPT rules for the specific IPs and ports you want to allow, and add DROP rules to block everything else.
Rules are evaluated top to bottom and the first matching rule wins. The usual pattern is to put specific allow rules at the top and a broad deny at the bottom; changes take effect within at most one minute. For detailed configuration, see the Firewall page.
Billing
On-Demand
You are billed only for the time a VM is running, with no commitment. While running, on-demand VMs are not reclaimed due to resource shortages (unlike spot). While stopped, you are not billed for compute, but attached storage and public IPs continue to be billed.
Reserved
A separate sales contract under which you reserve resources for a fixed term. The reserved resources are guaranteed for the duration of the term, and you use them at a discounted, stable unit price.
Spot
A discounted rate on idle GPU instances, relative to on-demand. The spot rate applies only to the GPU instance type; other attached resources such as block storage, public IPs, and networking are billed at standard rates. The discount fluctuates with supply and demand.
Monitoring
Metric
Numerical data collected over time about the state of a resource such as a VM or storage volume. Examples include GPU utilization and VRAM usage, CPU utilization, memory usage, network traffic, and disk I/O. You can chart metrics over time in the Metrics Explorer or use them as threshold conditions in Alerts.
Alert
A feature that automatically sends a notification when a metric crosses a threshold you have configured. Common uses include detecting training completion, preventing idle GPU billing, and warning before a disk fills up. An alert is made up of three components:
- Alert Rule: defines which metric, under which conditions, should trigger an alert
- Alert Action: defines who is notified and how when an alert fires (email is currently supported)
- Alert Event: the history of alerts that have fired
For detailed configuration, see the Alerts page.
Security and administration
Organization
The top-level unit of an ECI account. An organization contains multiple users and resources (VMs, storage, networks, and so on), and resource isolation, billing, quotas, and role-based access control (RBAC) are all managed at the organization level. Organizations are identified at sign-in by the Organization Identifier.
Quota
The resource ceiling assigned to an organization. Quotas apply to every resource type — VMs, block storage, public IPs, virtual networks, and so on — and you cannot create resources beyond the quota. Contact the support team if you need an increase.
Access Token
A key that authenticates a user when calling the ECI API. By using a token in place of a password, external tools can access ECI; this is what you use to drive ECI from Terraform, CLIs, CI/CD pipelines, automation scripts, and anything else outside the portal.
Role-Based Access Control (RBAC)
A model that bundles permitted actions into roles and assigns those roles to organization users. Because you assign a role instead of granting permissions directly to each user, people doing the same job receive a consistent set of permissions quickly.
Four built-in roles are provided — Owner / Contributor / Supporter / Viewer — and you can also define custom roles for your organization's workflow. The permission-code format and detailed behavior are covered in the Role-based access control (RBAC) guide.
Audit Log
A record of every action taken in the portal (VM creation, deletion, configuration changes, and so on). Each entry includes the timestamp, user, action, and resource, and you can inspect the full detail in JSON form.
Notice
Announcements Elice sends to ECI users. Notices cover operationally important changes — service maintenance, new features, pricing changes, security advisories — and appear on the portal home screen as well as in the Notices menu.
Next steps
- ECI v2: diagrams of how resources relate and core concepts
- Pricing model: comparison of spot, on-demand, and reserved