# USE CASES


Dedicated Cloud

dedicated servers

dedicated servers

vRack

Load Balancing IP

storage

HA

scalability

database cluster

Ceph cluster

monitoring

message queue


Wimi: an infrastructure with high availability combining Dedicated Cloud and dedicated servers through vRack


Wimi is a collaborative platform launched in 2012
The founders of the French startup Wimi: Antoine Duboscqet and Lionel Roux
Overview of a WIMI shared calendar in Wimi (at top, shared Tasks)
Overview of project documents within the Wimi interface.
Overview of "AirTime" functionality (videoconferencing directly in the web browser with WebRTC technology).

The online collaborative work platform was launched in 2012. Today, it has been adopted by more than 25,000 organizations. It wishes to facilitate teamwork by offering an interface that incorporates all essential services for effective collaboration: synchronisation and document sharing, (Wimi Disk), task scheduling, calendar sharing, chat, screen sharing, video and audio conferencing, (Wimi Air Time)… All associated services include management with appropriate access rights for professional use. To host this platform, the French startup has built at OVH an infrastructure that is highly available, combining Dedicated Cloud and dedicated servers through the use of a vRack private network.




Key information


  • 70 servers: 8 physical servers + 63 VM running on OVH.com's Dedicated Cloud (4 L Hosts) as well as a locally hosted OpenStack cluster (connected to the OVH infra via VPN)
  • 60 000 worldwide users
  • 1 000 https requests per second
  • 4 000 SQL requests per second
  • 10 GB database
  • Storage capacity 33 TB (files replicated 3 times)

Specifications


  • High availability of service: the application must be available 24/7/365
  • Data security: objective- zero loss of data
  • Horizontal scalability: it is necessary that the infrastructure evolve in a transparent manner and adapt accordingly to the number of connected users
  • Data hosted within French territory: constraint imposed by certain businesses/organizations and a marketing argument in a sector where for the most part the major players are American
  • Performance: interest is low in an application that lacks fluidity compared to local use ( = > need for a cache system)
  • Must provide users a secure connection (https) and an infrastructure for administrators (VPN two factor authentication)




Solution deployed at OVH





The high availability of the infrastructure is based on several elements. First is the choice of a native HA solution that Dedicated Cloud provides which includes redundancy of each component (hosts, power supply, and network connection). Second, is the separation of the group of services. Each service is provided by a cluster of independent servers, guaranteeing resiliency of the infrastructure. Within each cluster, a machine malfunction (or even multiple machines simultaneously) has no impact, functioning servers in the cluster temporarily absorb the additional load. In addition, this configuration also guarantees the scalability of the infrastructure, having the ability to augment the capacity of the infrastructure by deploying additional machines at the level of the points of congestion. This is an operation simplified by the utilization of Puppet (to be replaced by SaltStack in the future) to automatically deploy and configure additional virtual machines. To improve infrastructure scalability even more, the project evolves from a model of essentially polling (checking for updates at regular intervals) to a push model with the use of technologies such as WebSocket + node.js. That is to say that content is pushed to the user, rather than waiting for browser queries to reload the page or for java script to execute AJAX requests.

1 – Load distribution with 2 IP Load Balancing

2 - Front-end (Nginx + PHP-FPM)

3 - Back-end/Cache (PHP-FPM + Memcached)

4 – Data Bases (Galera cluster)

5 – File Storage (Ceph)

6 - Message queue (Rabbit MQ)

7 - Monitoring and performance analysis

8 - Backups

9 – The infrastructure's internal network



Feedback


“In 2012, when we were looking for a hosting provider to build our Wimi infrastructure, there were few choices available to us. Only the major players were offering virtualized infrastructure solutions combining flexibility and performance. But these players were American and we wished to reassure our users about the confidentiality of their data. We discovered OVH.com and their Dedicated Cloud offer during this time. In addition to the delegation of hardware management, the advantage of Dedicated Cloud lies in the fact that resources are dedicated and hosts and datastores can be added as needed. The arrival of the vRack was the decisive factor. It allowed us to combine the best of both worlds, the cloud for flexibility and dedicated servers for raw power. This is especially useful for hosting high volume databases. Without this private network, it would have been impossible to implement such technical choices such as the Galera and MariaDB clusters or the Ceph distributed storage. The interconnection of the machines would have been possible via their public IP, but that would have affected the overall performance of the infrastructure and raised safety concerns. Our ambition for 2015 is to develop Wimi internationally and in particular in the Americas. We’re investigating the possibility of extending/duplicating our infrastructure in OVH.com’s Canadian data center BHS.”

Thomas Lemarchand (@t_lemarchand), is Wimi’s information system manager.



OVH advice


The infrastructure deployed by Wimi is quite interesting as it combines all of the following: hybridization of virtual and physical resources assuring high availability (HA) and scalability, the adoption of OVH.com’s latest innovations (Wimi was among the first users of Dedicated Cloud, vRack, IP Load Balancing service, FS-Max servers...) and the use of the most advanced technologies (Ceph storage, MariaDB and Galera clusters, monitoring ELK ...).

One recommendation would be to set up a mechanism to monitor and identify critical thresholds for automatically adding additional resources. This way, when you pass a load threshold for a predefined amount of time, you would be able to start and stop scripts on new VMs via the OVH API. The VMs would be configured via Puppet or another configuration management system, and then automatically integrated in just a few minutes into the clusters that need reinforcements. Nevertheless, Wimi is a frequently used work tool and load peaks can be predicted. The infrastructure, which is nearly entirely automated, can therefore deploy an additional resource in 30 minutes - a very generous timeframe. With that in mind, the mechanism isn't absolutely essential.

On a final note, static web application files (JavaScript and CSS codes) would be better stored on the CDN. This would reduce the burden on front-end servers while improving the rendering time of web pages, even if they represent a minor share of global queries on infrastructure.



Elements of the infrastructure




1 Load Balancing IP

$13.00/month

Order


Dedicated Cloud

Enterprise Pack L 2 hosts AMD + Storage 2 x 300 GB

$782.60/month

Additional on-demand resources

Order


vRack

vRack
Service included with the Dedicated Cloud and the Infrastructure range servers


Databases
(Galera Cluster)

3 Infrastructure range servers
Model EG-32

$115.00/month

Order


Store files with Ceph

3 Storage range servers
FS-MAX (Pay-as-you-grow)

$403.99/month

Extra disks

Order


Backup

2 Storage range servers
FS-72T Model

$571.99/month

Order