Rapid.Space Hyper Open Cloud can deliver Hyper-Converged Infrastructure on-premise.
We will introduce in this presentation what is Rapid.Space, what is Hyper Open Cloud (open source, open hardware and open service) and how this translates into a transparent and cost efficient solution for Hyper-Converged Infrastructure.
Rapid.Space has two web sites: https://rapid.space (available worldwide except in mainland China) and https://rapidspace.cn (available in mainland China).
The primary service of Rapid.Space is a high performance virtual private server (VPS) at reasonable cost, combined with a CDN infrastructure for accelerated web content delivery.
Rapid.Space was founded in 2020 by Nexedi, Amarisoft and a few VIPs from IT and telecom industries.
Nexedi brings to Rapid.Space its open source stack, in particular its billing platform, its edge-cloud platform and its big data platform, all open source.
Amarisoft brings to Rapid.Space its purely software defined 4G/5G stack which covers all aspects needed for commercial deployment, including SA, NSA, NBIoT, etc.
Rapid.Space's concept is to provide to developers the minimum they need in order to deploy an application worldwide.
There are three basic services: Virtual Private Server (VPS), Content Delivery Network (CDN) and Software Defined Network (SDN).
VPS provides a way for developers to install their applications. It is similar to dedicated server services from companies such as Scaleway or Hetzner in Europe.
CDN provides a front-end solution to deliver data to end-users or to collect data from IoT. It is similar to Cloudflare or qiniu CDN in China.
SDN provides a way to interconnect Rapid.Space CDN and VPS through a latency-optimized IPv6 network. This service is quite unique: it also provides a way to interconnect Rapid.Space to other cloud services (AWS, Azure, GCP, Scaleway, Alicloud, UCloud, Qingcloud, etc.) with good networking performance.
Based on this minimal approach, developers should install by themselves open source software on VPS and build their applications: database (MariaDB, PosgreSQL, MongoDB, etc.), web server (Apache, Nginx, etc.), load balancer (Haproxy, ProxySQL, etc.). They should rely on the vast libraries available in python, PHP, ruby, Java, golang, nodejs, etc. to extend features.
Developers may use whichever tool they prefer for devops: SlapOS, OpenSVC, Docker, Kubernetes, Ansible, Chef, Puppet, buildout, etc. Even though Rapid.Space is based on SlapOS and buildout, Rapid.Space service can be used with other devops technologies.
The philosophy of Rapid.Space is thus the opposite of conventional cloud providers in the USA or in China. Rapid.Space provides very few services and lets developers rely on open source to achieve their goals. Thanks to this approach, developers can keep control on their applications and later move, if they wish, to another cloud platform.
There is no vendor lock-in.
Some types of services can be difficult or time-consuming to implement for developers on their own. This is the case of services that require clustering (use of multiple servers), hard real-time (industrial edge) or radio frequencies (4G/5G vRAN).
For each of these services, Rapid.Space provides a solution based on open-source software.
Rapid.Space provides a "Big Data" platform that combines the features of a data lake with transactional object storage, high-availability scalable relational database and out-of-core data processing in python (AI, physical models).
Rapid.Space provides an "Edge" platform that is optimized for automation (factory, building, etc.) and remote deployment of AI models.
Rapid.Space provides a "vRAN" network management system suitable for 4G/5G private networks (factories, hospitals, etc.) or public networks (telecom, government).
All advanced services are available to selected B2B customers as preview. General availability is expected in 2021.
All services are provided with source code under open source license (Big Data, Edge) or business license (vRAN).
Read online: How does Rapid.Space and SlapOS compare to AWS?
Any service that does not fit into Rapid.Space basic services (VPS, CDN, SDN) or advanced services (Big Data, Edge, vRAN) can be developed as a custom service.
Based on an early assessment, 85% of cloud services provided by Amazon AWS could actually be implemented with Rapid.Space low cost, high performance cloud and the various open source stacks such as SlapOS (75% services) and a few third party Free Software (10% services).
Rapid.Space provides a Platform as a Service (PaaS) so that developers can add new services to Rapid.Space.
Server-based custom services are developed with buildout language and SlapOS nano-container technology. They cover features such as:
A collection of sample buildout profiles is provided. They cover a wide range of cloud services and even include an open source ERP.
One should however keep in mind that many cloud services are actually no longer required with the introduction of technologies such as Progressive Web Applications (PWA). Quite often, there is even no need to develop a custom cloud service for Rapid.Space. A PWA will do better. This is due to the fact that a lot a server based architectures can now be implemented as browser based. Not only this saves time, money, energy and CO2, it also provides better scalability and portability.
Rapid.Space is available in Europe (France, Germany, Sweden, Nertherlands, Bulgaria), in Shanxi (north of mainland China) with two data centers and in Taiwan.
Rapid.Space IPv6 backbone is based on a hybrid mesh network which relies on hundreds of routers worldwide. Thanks to babel technology (RFC 6126), all sorts of congestions can be avoided. Latency can be minimized.
Rapid.Space provides HTTPS front-ends (HTTP1, HTTP2, HTTP3) in 10 different locations worldwide. In China, Rapid.Space front-ends are placed with all major carriers: CT, CU and CM.
Rapid.Space is supported by a network of experienced cloud integrators. Rapid.Space selects cloud integrators and engineers with a long experience in at least 3 of the open source core technologies used by Rapid.Space.
Rapid.Space cloud integrators can provide training, cloud migration services, operation management automation, big data lake implementation, 4G/5G vRAN deployment and industrial edge computing.
Current network covers most of the European Union, Russia, Japan, mainland China, Taiwan, Argentina and Brazil. We expect it to extend it soon in North America and Africa.
The goal of Rapid.Space is to provide sovereignty and trust through full reversibility. You may consider this goal as providing the kind of service that companies such as Huawei, Palantir or AWS are not able to provide due a combination of IP and legal policies.
This goal applies to every business which Rapid.Space is targeting.
Rapid.Space already provides a reversible cloud platform that can be used for public or private clouds. All components of this platform are open source, including the hardware, meaning that any customer can "clone" this platform on-premise or have it operated by a third party at no license cost.
Rapid.Space intends to provide a reversible big data platform with a scope similar to Palantir. All components of this platform are open source, including the hardware, meaning that any customer can "clone" this platform on-premise or have it operated by a third party at no license cost.
Rapid.Space intends to provide a reversible Edge computing platform which includes everything needed for Industry 4.0, including PLC, sensors, actuators. Again, all components are open source.
Rapid.Space intends to provide a reversible RAN platform which supports 4G/5G and can be used for both private and public networks. Most components are open source. Some components may be licensed source, meaning that any customer can "clone" this platform on-premise and audit its source code at some license cost.
Rapid.Space software is open source or licensed source except for a few exceptions.
Nexedi's SlapOS provides the OSS/BSS and edge cloud stack.
Nexedi's re6st provides the latency optimised hybrid mesh.
Accton's OpenAOS provides the network operationg system of switches.
Amarisoft stack provides 4G/5G vRAN stack.
Let us have a closer look at what Hyper Open Cloud means in practice.
For example, in Rapid.Space one can purchase online the same hardware as the one used by Rapid.Space in their own data centers. Rapid.Space tells you which model is used from which supplier. Most of the hardware are open source. A few of them are licensed source, which means that it is possible to access the industrial design source files but under a non open source license.
The right to study a Hyper Open Cloud takes, in the case of Rapid.Space, the form of a handbook and of public audits.
Rapid.Space handbook not only describes how to use the cloud service. It also describes how to operate a point of presence, step by step. Management processes of Rapid.Space International company are also being added little by little to the handbook, as in a work-in-progress through constant updates.
Rapid.Space is based on an open source technology called "SlapOS" which is quite different with other open source cloud technologies.
Most solutions (Kubernetes, OpenNebula, OpenStack, OpenSVC, Proxmox, XCP-NG) have their strengths but are overall insufficient to build a complete cloud per se . Many features are missing. Much integration effort or custom development is required to cover those missing features.
If one needs a proof of this situation, simply observe OVHCloud, the largest cloud operator in Europe and one of the top-10 cloud operators in the world. Despite the fact that OVHCloud operates already both OpenStack and Kubernetes clusters, the company ended up purchasing a license of Anthos software from Google to cover the features that were missing.
What is needed to build a cloud in addition to virtualization software such as qemu/kvm or orchestration abstractions such as containers or nano-containers, is what people call an OSS/BSS software in the telecom industry. The term OSS/BSS stands for operation support system and business support system. It is a kind of cloud ERP which encapsulates operation management, orchestration, disaster recovery, monitoring, lifecycle management, issue tracking, accounting and billing.
Rapid.Space is built around one of the only open source cloud software which supports all the OSS/BSS requirements for a cloud. It provides a service profile format to specify what is a "cloud service" by encapsulating self-healing, monitoring, accounting, disaster recovery, portability, system upgrade, resource sharing and delegation in addition to basic build and run.
In addition to OSS/BSS, a complete cloud service requires tight integration with the network operating system (NOS) down from the top-of-rack (ToR) switch up to the global end-points of the CDN. This is also something which is rarely covered by most FLOSS solutions besides SlapOS and Rapid.Space.
Yet, even though SlapOS covers all those OSS/BSS features, it is still not enough. Without formalized operation management procedures to train and organize staff, it takes about two years to acquire the know how and operate a cloud.
Rapid.Space storage performance is another distinctive advantage. Thanks to direct attached storage, both Rapid.Space PaaS and VPS can achieve transactional database performance much higher and at lower cost than solutions based the use of virtual block device.
Storage performance is actually one of the most common problems with cloud and Hyper-Converged Infrastucture (HCI). Many early adopters have faced "bad surprises" with write-intensive database applications. This problem has been addressed in order to meet the requirements of a very larger ERP programs.
The concept of Hyper-Converged Infrastructure with Rapid.Space is based on three steps: purchase, connect and operate.
Rapid.Space sells open source hardware similar to the one used by Facebook, Twitter, Yahoo! Japan, etc. This hardware was designed by the "Open Compute Project" (OCP). It includes servers and switches.
Once our customers purchase the hardware, they connect it by themselves in their data-center. Rapid.Space OCP servers are compatible with both 19-inch and 21-inch rack standards. Each group of servers is powered by a single powershelf, which saves electricity and simplifies maintenance.
Rapid.Space provides installation instructions for the software that operates servers. It takes about 5 minutes to setup one server (a bit longer for the first one).
After all servers are up and running, users can connect to Rapid.Space panel and request services on their servers.
Everything is included in the price of each Rapid.Space server: operating system, operation support system, network operating system, infrastructure as a service (IaaS), platform as a service (PaaS), software defined network, content delivery network. All software are open-source.
Every server is actually provided with a lifelong "Rapid.Space token" which ensures that the server will be automatically maintained, updated, monitored and operated through Rapid.Space software technology and service.
Rapid.Space provides remote installation and support of the servers thanks to smartglasses. Every customer is provided with a RasberryPi which is used to remotely diagnose and maintain a server. Whenever the RasberryPi is disconnected, Rapid.Space team does not have access to the servers thanks to the "Zero Knowledge" privacy option.
Rapid.Space provides 3 models of servers and one model of switch that have been certified to operate reliably with Rapid.Space software and services.
The price of servers is roughly equivalent to the factory price plus the "Rapid.Space token" price.
Thanks to the permanent token, servers can be operated with Rapid.Space software and technology at no extra cost, until their end-of-life.
Rapid.Space also provides the option to rent servers through 36-month (Leopard) or 60-month (Tioga Pass, Capri, AS5812) contracts.
Hyper-Converged Infrastructure with Rapid.Space is a matter of options.
Sadly, there is no "one size fits all" solution with cloud due to the constraints imposed by storage. If one adopts "High Availability" (HA) technologies, then the risk of disaster increases. If one adopts automated disaster recovery (DR), then big data is no longer an option. If one goes for PaaS, then HA, DR and performance are possible but it takes some time to port applications.
The best solution is thus a matter of trade-off between HA, DR, performance, flexibility and learning curve.
For applications that do a lot of writes to the storage and that do no need system-HA or system-DR, Rapid.Space recommends a Standard VPS.
For applications that do a lot of writes to the storage and that need system-HA, Rapid.Space recommends a Redundant VPS which consumes twice as much resources at least.
For applications that may require DR but do not do a lot write to the storage, the "Resilient VM" is ideal and very flexible in terms of resource allocation or over-allocation.
For maximum performance and flexibility, "Rapid.Space PaaS" is the best option but requires some training.
By default, the recommended networking topology uses IPv6 on the backend and IPv4 on the frontend. Other topologies are possible. However, we found that using IPv6 on the backend provides more flexibility and scalability while using IPv4 on the frontend provides more interoperability.
Developers are thus encourage to run IPv6 on their development platform through native IPv6 of their ISP or through an IPv6 interoperability layer such as teredo, miredo, Rapid.Space SDN, etc. Developers will usually access VM, VPS or PaaS instances over IPv6.
Even if backend servers are located in a single location or in a few locations, proxies can be located in multiple places around the world in so-called "edge POP". This approach is similar to what Facebook does for its own infrastructure. It makes web sites more reactive by reducing the time to establish an HTTPS connection for the frontend and by reducing the latency between the frontend and the backend. It also increases resiliency by having more than one frontend to serve the end-user and by having more than one route between the frontend and the backend.
All the services that we described (VPS, VM, PaaS, CDN, SDN) are currently used by Nexedi for its own operations. Nexedi is one of the largest open source software publishers in Europe. It provides mission critical applications to Airbus, PSA Group, Toyota, SANEF, Kyorin, etc.
All developer instances of Nexedi are hosted in PaaS or VMs.
Some production instances are hosted in VPS while others are hosted on 3rd party hardware, either on a public cloud or on customer's premise. This demonstrates the ability for Rapid.Space technology to operate in a multicloud context.
Nexedi is using 4542 domains on its HTTP proxies. Most of them are configured automatically.
Rapid.Space Hyper-Converged Infrastructure can be deployed on-premise either by connecting to Rapid.Space's central orchestrator (Standard) or by deploying a dedicated orchestrator (Dedicated).
With "Standard" deployment, servers deployed on-premise form a hybrid cloud infrastructure with the public infrastructure of Rapid.Space. Cloud loads can be deployed on-premise, on the public cloud or a combination of both. Nexedi (see above), is using this "Standard" model.
With "Dedicated" deployment, servers deployed on-premise form an isolated herd of servers, independent of the Rapid.Space public infrastructure. Teralab (see below) is using this "Dedicated" model.
It is also possible with "Dedicated" setup to allocate a dedicated range of IPv4 and IPv6 adresses or integrate Rapid.Space with custom networking architecture.
Rapid.Space's SlapOS technology has been deployed by Teralab in France. Teralab is one of the founding member of Gaia-X, the European cloud project. Teralab uses a "Dedicated" deployment with extra security rules designed to protect sensitive data through 3 levels of access control.
Teralab is mainly used for big data applications.
Rapid.Space partners provide professional services to ease the adoption of cloud technology and hyper converged infrastructure. The table of services presented here provides indications of price. Price may however vary from one partner to another.
The most important service is "onboarding". It is based on the standard lectures of Rapid.Space. For each lecture, group is created on a messaging application (Telegram, WhatsApp, Wechat, Signal, etc.). A plan is defined to drive the user towards understanding and adoption.
Each step of the lecture is implemented by the user, if possible in the context of the kind of applications that he or she is planning. At the end of the "onboarding" process, the user understand how to use Rapid.Space technology and why it was designed in this way.
In addition to onboarding, Rapid.Space partners provide services to support each user. If a user wants to ask a third party to "create a VM with pre-installed OS", just ask Rapid.Space partner and someone will do the installation for you. If there is any question, just ask and someone will find out. This is what we call "managed instance".
P2V (physical to virtual) and V2V (virtual to virtual) migration services are here to help you migrate existing computers and virtual machines to Rapid.Space technology. Partners will the use different tools depending on the case. If a lot of migration is required, some automation is possible.
SSL and ICP domain installation is the service where a Rapid.Space partner sets up a CDN with a custom domain after requesting all necessary information and handling bureaucracy. SSL domains are secure domains outside China. ICP domains are their equivalent in China, with the mandatory steps to register a domain to local authorities.
Rapid.Space partners can develop or help developing buildout profiles. For example, if a partner wants to provide Nextcloud instances automatically, Rapid.Space can create a buildout profile to automate Nextcloud deployment and operation. The cost is based on time and material.
Rapid.Space partners can also configure a SlapOS master and help maintaining it. This includes updating the list of services delivered by the master to SlapOS nodes, delete some users, etc.
Zonova is a company doing Web applications and mobile applications. They use Portainer to manage flocks of containers. Rapid.Space onboarding focuses only on one lecture: VPS, SDN, CDN (basic usage). After a few hours spread over a couple of days, Zonova could deploy its infrastructure on Rapid.Space VPS, by themselves, based on the tutorials and instructions provided by Rapid.Space partner.
For more information, please contact Jean-Paul, CEO of Rapid.Space (+33 629 02 44 25 or email@example.com).
Rapid.Space publishes guidelines online with all steps to make setup easy. Read for example: HowTo Physically Install ESA Kit and OCP Servers.
Based on our experience, an inexperienced technician can achieve all steps after carefully reading instructions.
Regarding software setup, Rapid.Space provides remote assistance to setup operating system through a local RaspberryPi with a custom system image. All steps are also published online.
We recommend the use of IPv6 for backend for many reasons.
First, anyone can easily purchase their own IPv6 range, and thus control their own network entirely instead of depending on someone else. It is cheap and easy. This is no longer possible with IPv4.
By using IPv6, we can allocate a complete range to each server. Each server thus a million or more IPv6 adresses. Each process in a server can have its dedicated IPv6. There is no problem of port conflicy anymore.
IPv6 has auto-discovery. This makes it possible to automate the configuration of a network. This is actually what re6st does (re6st is the technology of Rapid.Space SDN). By discovering automatically which server connects to which other server, and by doing this recursively, the configuration of the entire worldwide network of Rapid.Space servers is automated, made resilient and optimized for lowest latency.
Also, IPv6 removes NAT. All processes have a global IPv6 which is reachable from all other processes. The addressing is thus flat, which makes orchestration and distributed computing much easier. Firewalling is possible of course to filter packets and decide who can reach who.
IPv4 can be used with Rapid.Space in different manners: outgoing (to reach other sites outside the infrastructure), incoming (to be reached by IPv4) and proxied (to be reached through an IPv4 to IPv6 proxy).
It is also possible to assign an IPv4 range to each server as part of "Dedicated" options. It will then become possible for services deployed on Rapid.Space HCI to interconnect over IPv4. This is what Teralab does.
However, we do not recommend to use IPv4 currently because it leads to "bad development practices". HADOOP is a terrible example of such evil practices: instead of using the libraries designed to handle IP addresses in a generic way, many HADOOP libraries are using regular expressions to parse IPv4 addresses. This obviously crashes on IPv6.
Having an "IPv6 only" backend infrastructure is actually a good way to ensure that developers and companies are enforcing compatibility with IPv6.
Rapid.Space technology automates the process of backup (ie. make a copy of data), archive (keep a history of backups), restore (rebuild an system from its source code and data archive) and test (ensure that the rebuilt system behaves in the same way as the original one).
This automated process can be launched every day to every hour.
It is available for PaaS and VM (not for VPS).
By combining this process with a distributed architecture (VM, PaaS) on multiple sites, a resilient network (Rapid.Space SDN) which can circumvent routing incidents and multiple proxies (CDN), it is possible to reduce drastically the risk of disaster and address the problem of disaster recovery in a deterministic, rational way.
Rapid.Space can provide HA but recommends to focus first on DR. Too many companies with HA infrastructure end up losing everything due to the absence of proper DR.
The CAP theorem states that it is impossible for a distributed data store to simultaneously provide more than two out of the following three guarantees:
VM and PaaS services in Rapid.Space focus on Consistency and Partition tolerance but not on Availability. In addition, PaaS provides highest performance, but not virtualization.
VPS service in Rapid.Space focuses on Consistency but not on Availability or Partition tolerance. VPS provides both virtualization and performance.
VPS HA service (Q2 2021) will focus on both Consistency and Availability but not on Partition tolerance. VPS HA is expected to provide both virtualization and performance.
In addition to VPS HA, it is possible to achieve HA at database level: RepMan for MariaDB, NEO for ZODB, etc.
The price of a typical implementation of a Rapid.Space HCI environment depends on the hardware and services which are used.
If we consider a 19-inch rack setup, the first step is to purchase:
This leads to an initial setup which can later be extended to 16 servers per ESA Kit and at most 2 ESA Kit per 19-inch rack.
Setup of the rack can be done through remote assistance by Rapid.Space, or by asking local partner.
Rapid.Space local partner provides onboarding for the initial use (VPS, CDN, SDN) and for more specific use (VM, PaaS).
Rapid.Space local partner provides migration service for some VMs and some physical servers.
Rapid.Space local partner provides managed instance service to help users operate their first VPS, VM or PaaS.