Hello, we've been using Zabbix 1.4.4 for ages. We're eager to migrate to a new version for various reasons. We started with MyISAM table structure, but the new patches provide new tables in InnoDB table format. That would require MySQL to support both MyISAM engine as InnoDB engine.
Our current server only has 4GB of memory (32bits OS), and from what I've been reading, an InnoDB structure strains both CPU and memory more than a MyISAM structrure, at the gain of row level locking instead of table locking.
Of course I can use mysqldump and sed to change the entire structure over to InnoDB, but dumping and loading takes ~15 hrs to complete, at the risk that at the end, the InnoDB structure is straining the available CPU and RAM resources to the max. I guess the only way to speed up this 15 hrs, is to do it on new hardware, but I'd like to see the extend in which Zabbix 1.8.3 is able to reduce server load.
So, I'd like to stick with the MyISAM structure for now, use sed to change "InnoDB" to "MyISAM" in the DBpatches 1.4.4 -> 1.6.x -> 1.8.3, see how that performs, wait for about a month to see how things are running, and then get some budget for a new server, 64bits, with lots of memory, and do the DB engine migration on new, faster hardware.
But I wonder (and here's my question
), if it is wise to do the conversion of the patches as I described, or will that decision bite me in the *ss?
For example, I can imagine that those patches are specifically designed with the InnoDB row lock feature in mind, I'd like some input before I endeavor into my demise
<edit> current DB size is 8,2GB, converted to 1.8.3 + MyISAM->InnoDB conversion it's about 17GB.
If I knew how to upload attachments, I would upload a "show processlist", but in short it's displaying 62 "insert into history" statements, 5 selects on history, and about 56 sleeps
Our current server only has 4GB of memory (32bits OS), and from what I've been reading, an InnoDB structure strains both CPU and memory more than a MyISAM structrure, at the gain of row level locking instead of table locking.
Of course I can use mysqldump and sed to change the entire structure over to InnoDB, but dumping and loading takes ~15 hrs to complete, at the risk that at the end, the InnoDB structure is straining the available CPU and RAM resources to the max. I guess the only way to speed up this 15 hrs, is to do it on new hardware, but I'd like to see the extend in which Zabbix 1.8.3 is able to reduce server load.
So, I'd like to stick with the MyISAM structure for now, use sed to change "InnoDB" to "MyISAM" in the DBpatches 1.4.4 -> 1.6.x -> 1.8.3, see how that performs, wait for about a month to see how things are running, and then get some budget for a new server, 64bits, with lots of memory, and do the DB engine migration on new, faster hardware.
But I wonder (and here's my question
), if it is wise to do the conversion of the patches as I described, or will that decision bite me in the *ss?For example, I can imagine that those patches are specifically designed with the InnoDB row lock feature in mind, I'd like some input before I endeavor into my demise
<edit> current DB size is 8,2GB, converted to 1.8.3 + MyISAM->InnoDB conversion it's about 17GB.
If I knew how to upload attachments, I would upload a "show processlist", but in short it's displaying 62 "insert into history" statements, 5 selects on history, and about 56 sleeps
Comment