We went further than we planed in the WM architecture optimization phase.
In fact, 5 persons have worked for a month on the improvement of the display speed of your sites and management interface.
In fact, 5 persons have worked for a month on the improvement of the display speed of your sites and management interface.
The starting point: an increase of the Web site traffic.
A month ago, we had a huge increase of the website traffic: +35% pages. We hadn't planed such an raise within two weeks.
Back then, we already had bought 8 new servers extremely powerful. They were last generation Bi_Xeon QuadCore, equiped with 16Go Ram, and with the best SATA disks available: Western Digital's Velocity Raptors.
Unfortunately, this gain was to be quite elusive, considering this sudden traffic raise. So we challenged ourselves, optimizing the application code, but we didn't know where to start from.
Back then, we already had bought 8 new servers extremely powerful. They were last generation Bi_Xeon QuadCore, equiped with 16Go Ram, and with the best SATA disks available: Western Digital's Velocity Raptors.
Unfortunately, this gain was to be quite elusive, considering this sudden traffic raise. So we challenged ourselves, optimizing the application code, but we didn't know where to start from.
The approach:
Each WM developer has made a list of parts to improve, that could impact the display speed.
We've stagnated for 2 weeks. The advancements were modest and we were about to have our architecture checked by sun engineers... but the work we've operated on a new wrapper SQL finally payed of.
Indeed, we've been working on an abstraction layer of the data base for a few months. Among other things, this system provides a monitoring of the 3000 requests/s operated on the 10 WM SQL servers. Thus we can trace in a control table, the requests the most frequent, the longest, etc...
Having tired out all ideas to optimize the response time, we rewritted, thanks to the control table information from the wrapper, a dozen SQL request a day for a week, and we reached unexpected results!
We've stagnated for 2 weeks. The advancements were modest and we were about to have our architecture checked by sun engineers... but the work we've operated on a new wrapper SQL finally payed of.
Indeed, we've been working on an abstraction layer of the data base for a few months. Among other things, this system provides a monitoring of the 3000 requests/s operated on the 10 WM SQL servers. Thus we can trace in a control table, the requests the most frequent, the longest, etc...
Having tired out all ideas to optimize the response time, we rewritted, thanks to the control table information from the wrapper, a dozen SQL request a day for a week, and we reached unexpected results!
Gains: response time 3 times improved.
Thus we decided to continue. Today, 3 weeks after, not only have we totally absorbed the traffic raise, but we've also improved the response time by 3!!!
Thanks to this work, we probably are one of the market fastest SaaS website platform.
Thanks to this work, we probably are one of the market fastest SaaS website platform.
And now?
To the end of January, we'll dedicate a day a week to this kind of optimization. We aim to reach a 250ms response time on average.
After this optimization month, we'll resume the publication of novelties. By the end of november, we'll reach again our cruising speed.
We're going to start with the tags
After this optimization month, we'll resume the publication of novelties. By the end of november, we'll reach again our cruising speed.
We're going to start with the tags
