OVH Community, your new community space.

RPS update ... go
23-02-2010, 10:14

3 weeks ago, we received a patch from Sun for
the ZFS. This patch fixes a major problem in the
management of our storage infrastructure for the RPS. Especially
at the time of service restoration after a reboot
a spin. Basically, after a crash, power failure or maintenance
the filer had around 2Hours of unavailability for the service.
This problem is now with R&D to find solutions
to a few problems such as the search performance of
the storage. Therefore, we have preferred to shield the current system
production and prevent failures that continue in R&D and
offer better performance and flexibility. Similarly
NFS storage in the RPS has been frozen. Basically, the fact
that the storage is considered slow (1MB/sec) comes from there.

With this patch, everything restarts.

We worked for 3 weeks rethinking the
infrastructure of the backup for all RPS filers. The problem
is that at this level they have been multiplied by x100: if a reboot spin
backup must wait 1 week to have access and
then repeat the backup. Here, with this patch, everything becomes viable
again. Work on infrastructure for the RPS backup is
almost finished. We will began work on
iSCSI filers shortly, which will take us a few more days.
Then we'll begin beta testing different ideas we have in
stock and those currently in the "Cloud" with all the great
positive feedback (we read your comments).

In the end, we should complete all these updates and
R&D at the start of the first RPS geolocation, first
in Paris (it's already in beta) but also in Frankfurt and Madrid.

The purpose is threefold:
- Propose a hardware platform for online games with
latency at the lowest level possible
- Implement geolocation infrastructure throughout our own
network but also in Japan, USA, Russia and Brazil with the objective
of service distribution of content to the visitor for
Web caching, streaming, or simply broadcast content
- Working on the concept of a server being "100% online", ie
a server that is never 'down' in any way. if there is
any problem or maintenance, a hot failover server
engages automatically which also includes storage, and
the service will know no interruption.

That is where we want to go with our technology which will be called "Father".
It is therefore likely that the RPS product in its current form
will be more controllable in 2-3 months. And the only place you
can order a complete turnkey solution, 1 click with simple
use for specific purposes with defined performance.
The RPS' in place will remain in place, of course. We assume that,
as long as the service is paid we will keep improving it.