RSS for all categories
FS#4393 — FS#8328 — Establishment of a new load balancer
Attached to Project— Hosting
| Modernization | |
| 1000gp | |
| CLOSED | |
 |
We have developed a specific load balancer , more accurate and a lot more flexible.
It has been specially designed to handle the shared hosting architecture.
Lab tests allowed us to improve the caches:
We have got up to 6x less disk queries.
We will deploy it tonight in order to confirm its performance in a real environment.
Remaining a few settings to make it optimal.
The load balancer is dedicated to work performed on okillerd (which reports any issue on your php scripts, on manager)
Date: Friday, 12 April 2013, 12:26PMIt has been specially designed to handle the shared hosting architecture.
Lab tests allowed us to improve the caches:
We have got up to 6x less disk queries.
We will deploy it tonight in order to confirm its performance in a real environment.
Remaining a few settings to make it optimal.
The load balancer is dedicated to work performed on okillerd (which reports any issue on your php scripts, on manager)
Reason for closing: Done
The load balancer was enabled.
The load balancer works properly.
We have many logs to analyze in detail.
We will stop it during the analysis.
Yesterday, we have validated the crash tests: if the system crashes,
the cluster continues to run and switches to the current repartition.
Thereafter the analysis of previous tests, we have improved
the prediction's algorithm.
We are going to activate the module this afternoon.
Load balancer is enabled.
We have reset the old load balancer for this weekend.
The distributor is working again and under monitoring's admins.
We will test the preduictor on start offer tonight.
We have enabled the predictors on cluster014 (the old start)
We are preparing the cluster015 (aka mailplan)
The predictor will be activated on Tuesday during the morning.
We have activated the predictor on the cluster 15 (mailplan)
The predictor is temporarily disabled on cluster 15.
Some sites use too many resources, we will have to isolate them on a dedicated one.
Thereafter diagnosis done on cluster 15, we have improved the repartition algorithm.
It has been deployed tonight on cluster 14.
We will reactivate the new predictor on cluster 15 on Wednesday morning.
The balancer is activated on cluster 15.
Balancers are validated on these clusters.
We will soon deploy them on other closed ones.