Some backstory: I run the servers for one of your clients. The client started on VB3, got a decent level of success, outgrew their dedicated server at Fast Hosts, by fate ran into me and the rest is history I guess. We've since upgraded to V4 a few years back and delayed moving to v5 in the hope that VB would sort out all the kinks. We're now on v5 and are running into some issues.
Performance wise; I find that v5 is using roughly 4x as much resources on the server. Disk IO usage is massively higher. I run the forum on a pair of HP servers, E3 CPU, RAID 10 SATA storage, plenty of RAM. We're running a hot spare setup. I ended up temporarily moving the forum to a RAID 10 SSD based server at Rackspace as the forum was swamping the IO on the server. The server is running some other small sites and email for a fairly large number of people. Prior to the VB upgrade server load was typically below 2.0 (on a quad core server). Are there any tips to impove the performance of the forum in general especially in regard of performing less writes to the filesystem or preventing MariaDB from creating temp tables on disk which is what appears to be swamping the servers IO?
My client is experiencing difficulty moving threads from one forum to another. On the production setup they are seeing errors related to the PHP processes going away when they attempt this action. Upon inspection I found that this is caused by PHP attempting to exceed its configured memory limit - so I increased it to 256MB, then 512MB, then 1GB, then decided to move testing away from the production server.
I copied the forum to a test bed server, NVME SSD, 32GB RAM, Quad Core 5GHZ CPU (yes overclocked). Setup mirrors the live setup - php5.6 mariadb10 nginx. I cleared the systemcache and optimised tables before starting.
When moving a thread into the main forum; which has about 1.25 million posts, I am seeing extreme resource usage on the server. After increasing various resource limits I see the forum attempting to run a 23+MB query (just the query text) on the database server. The query starts SELECT * FROM cacheevent WHERE `event` IN ('nodeChg_2426203'. After increasing the maximum query size on the server I see mysqld and php-fpm processes using around 3GB of RAM each. After about 2 minutes of CPU limited resource usage the forum thread is finally moved.
As you can hopefully appreciate the resource limitations required to get this to run far exceed what is workable for most shared environments. Indeed I will not be offering to increase resource usage limits to this degree on my shared cluster as this would allow this user to impact other clients. Hopefully we can optimise the performance of this operation somehow so that the client does not need to either go without the ability to move threads or upgrade to a very high end dedicated server that shall largely be sitting idle and still take several minutes to simply move a thread?
Performance wise; I find that v5 is using roughly 4x as much resources on the server. Disk IO usage is massively higher. I run the forum on a pair of HP servers, E3 CPU, RAID 10 SATA storage, plenty of RAM. We're running a hot spare setup. I ended up temporarily moving the forum to a RAID 10 SSD based server at Rackspace as the forum was swamping the IO on the server. The server is running some other small sites and email for a fairly large number of people. Prior to the VB upgrade server load was typically below 2.0 (on a quad core server). Are there any tips to impove the performance of the forum in general especially in regard of performing less writes to the filesystem or preventing MariaDB from creating temp tables on disk which is what appears to be swamping the servers IO?
My client is experiencing difficulty moving threads from one forum to another. On the production setup they are seeing errors related to the PHP processes going away when they attempt this action. Upon inspection I found that this is caused by PHP attempting to exceed its configured memory limit - so I increased it to 256MB, then 512MB, then 1GB, then decided to move testing away from the production server.
I copied the forum to a test bed server, NVME SSD, 32GB RAM, Quad Core 5GHZ CPU (yes overclocked). Setup mirrors the live setup - php5.6 mariadb10 nginx. I cleared the systemcache and optimised tables before starting.
When moving a thread into the main forum; which has about 1.25 million posts, I am seeing extreme resource usage on the server. After increasing various resource limits I see the forum attempting to run a 23+MB query (just the query text) on the database server. The query starts SELECT * FROM cacheevent WHERE `event` IN ('nodeChg_2426203'. After increasing the maximum query size on the server I see mysqld and php-fpm processes using around 3GB of RAM each. After about 2 minutes of CPU limited resource usage the forum thread is finally moved.
As you can hopefully appreciate the resource limitations required to get this to run far exceed what is workable for most shared environments. Indeed I will not be offering to increase resource usage limits to this degree on my shared cluster as this would allow this user to impact other clients. Hopefully we can optimise the performance of this operation somehow so that the client does not need to either go without the ability to move threads or upgrade to a very high end dedicated server that shall largely be sitting idle and still take several minutes to simply move a thread?
Comment