At the end of 2017, we started using a completely new solution for NoLimit. Everything has moved to the cloud and new servers, where efficient processors no longer mean a lower CPU frequency, but that they run and consume electricity according to the current required performance (they can underclock and overclock). In addition, a proxy server was built in front of the webservers that could automatically cache frequently used content.
Old and new NoLimit solutions
The difference between the old and the new solution was huge in terms of performance. Faster processors have been seen especially in content management systems like WordPress. Gradually, the configuration was tuned, a number of modifications and software were tested. Something had an effect on speed, something more on stability.
It may not seem like it, but we are constantly working on and improving the NoLimit service. For example, over the holidays, they moved all the site data to HPE 3PAR StoreServ Storage, which is storage that uses AI to optimize performance and predict problems. This storage can handle 3 million IOPS with a response time of under 1 ms. Hewlett Packard Enterprise guarantees data availability up to an incredible 99.9999%. They can scale up to 80 PB per system. The main reason was to make administration easier and more stable. However, we have also measured a small overall speedup, as customers who need to perform disk-intensive operations are relieved.
Of course, there were also a number of problems, which we gradually overcame and were able to move forward. While in the first year of operation we didn’t have 100% availability on any new server with NoLimit, because everywhere was always tuning something, it was the opposite with the old ones. If 100% availability was not maintained, the reason was either a planned outage (HW replacement) or the problem was somewhere else (network, attacks on infrastructure, etc.).
It’s easy to maintain 100% availability for services where nothing changes and they don’t move anywhere. But if you want to make progress and move things forward, you can’t do without downtime and sometimes outages. You know the saying, “He who does nothing can do nothing.”
How to migrate tens of thousands of services
We wrote about the migration to the new solution more than two years ago. There were two problems. Development was always evolving and colleagues were pushing the service and infrastructure further and further. Occasionally something fundamental has changed. The second problem was that everyone here assumes that everything must go without downtime. So when it came to designing a solution, it was very complicated, because there are a lot of things to think about in a failover migration. Moreover, web hosting is a complex service and the old and new solutions are quite different.
But the constant delays led to the old NoLimit becoming really old. Customers missed newer versions of PHP and what can we say, especially content management systems and their plugins are much more demanding today than five years ago. It was already showing in performance, especially in administration where you don’t have as many options to cache content.
So the company’s management banged the table, development was explained that a minor outage was fine and migration became the number one priority.
There were several test migrations during the holidays, but the full migration started after the holidays.
On October 14, 2021, the colleague in charge of NoLimit announced the successful completion of the migration.
In total, 45,000 services were moved from 127 servers with a total of 170 TB of data.
These were services from 2010 – 2017, so someone could have had something set up individually. Therefore, the support and CMS specialists were also tasked with reviewing all sites that had their status changed to WEDOS OnLine from these servers.
Yes, thanks to WEDOS OnLine we monitor all our customers’ websites ourselves. We measure both availability and responsiveness. Thus, we work not only with the data from the server, but also with whether the individual customer sites really work on them. Monitoring is carried out both from our datacenters and from the world.
This allowed us to reach out to individual customers and offer to help them with solutions.
By the way, WEDOS OnLine has also shown how much faster the sites have become on average after the move and the difference is really big. This is the TTFB (Time to first byte) value, i.e. the time it takes for the server to return the first data from the server to the browser.
Of course, not everything went smoothly
It was a really big event and everything did not go completely without problems. Apart from the problem with the individual settings of some customers’ websites, we gradually encountered other problems, which we promptly addressed and solved. The last current one was resolved on 20.10.2021 in the early morning hours.
So when something comes up, we don’t put it off and deal with it. We have the human capacity and experienced people to do it.
Since then, the servers are back to 100% availability. And now it’s true for all servers. New and old.
The future of NoLimit
We’re still counting on something coming up, so we’re on alert just in case. Everything is closely monitored and what you write to us is monitored.
When the situation calms down, we will face further challenges.
We still want to domigrate the databases to the 3PAR repository. Subsequently, we plan to add a secondary storage to each repository, which will be in a second datacenter and will “hold” a synchronous (on-line) copy of the data. So the data will be simultaneously written to 2 storage locations in different datacenters. This is a qualitative and preventive shift that is quite exceptional in similar services, even from a global perspective. At a time when everything is getting more expensive, we are investing and constantly developing the service.
Not to mention that the power supply in the second datacenter is from a hydroelectric power plant 🙂 the heat will be used to heat the town swimming pool and our office building. In the first datacenter we have our own cooling solution and so it is the most economical datacenter in the Czech Republic. The second is oil-cooled and it pushes the boundaries to even better levels. This is a unique worldwide event. So our hosting is green, like peas.
We have identified and even tested new technologies that could take NoLimit much further. However, some of them have paid licenses, so the question is whether we will go the way of an additional service or a completely new service.
We would like to add these new technologies to the WMS, where it will be easier. Licenses are usually tied to system resources and the WMS always has dedicated hardware. By the way, for WMS, this would bring the service incredibly forward for specific purposes. A million visitors would be no number. We know this because we were able to test it on a website that, according to official measurements, is one of the top ten most visited websites in the Czech Republic. Plus, most of his traffic is impact 🙂
A new and more powerful proxy solution is also being considered. In combination with the upcoming WEDOS AnyCast and WEDOS CDN service, we are no longer afraid of global challenges. It might all go faster than we originally planned 😉
We still have more improvements planned, but more on that next time.
Conclusion
So now everything is migrated and we can move on. All customers can now use for example PHP 8, faster processors, disks, proxy server and many other improvements we have made over the years with the new NoLimit solution.
Now we can finally move on. You definitely have something to look forward to 🙂
If you want to know more about NoLimit, the colleague who is behind its development gave a lecture at the last Ait’s Party.