At the moment the ones which will not be moving are backup, restore, and the plugin installer. We cannot move all operations, because some of them need state to persist between page loads. Reads are much faster though.Īs a side note, we are moving many of the operations which happen in tempdir to localcachedir where possible. The other thing to bear in mind is that RAID1 is typically slower than without RAID because you have to write to both disks and have confirmation that both disks wrote successfully before the operation completes. As a result it may be slower than if you are running in a single machine system. The test above is a write test of the tempdir, which is one of those directories which must be on a shared file system. The thing about shared file systems is that they are typically slower - they have to handle file locking, and are usually network based file-systems (which are also typically slower). The localcachedir is a local option which does not have to be shared. Of these, the dataroot, tempdir, and cachedir must be shared between all nodes in your cluster. These include the ability to set your dataroot, as well as tempdir, cachedir, and localcachedir. Moodle has several directory options available to you to help you with performance. There are a couple of things to bear in mind here. If you are worried about a tmpfs for your moodle tempdir being too small consider adding an SSD to your setup for the purpose instead, ramfs does not have space restrictions and will grow as large as free memory will allow will then spill over to swap causing a huge hit to your IO performance. but be aware of it and manage it if needs be.Īn alternative to tmpfs is ramfs. For you you have such a totally ridiculous amount of RAM this probably won't be an issue. If the backup process runs out of space it fails and can leave partial files lingering in your tmpfs taking up space causing knock on problems. This can need GBs of data depending on your course. Moodle uses it's tmpdir for all sorts of things, including backup and restore of courses. This is only used when data is stored, space allocated to tmpfs is available to the OS for other uses when it is not being used by the disk. Tmpfs by default allocates a size of 50% RAM to the drive. Yes, you do need to be aware of some things here. Google is your friend if you need help with any of the specifics here. Now this works your'll want to add these mountpoints to your /etc/fstab so that they come up automatically on reboots. Purge your moodle caches in the usual fashion and re-run the benchmark. you have bucket loads of RAM so I'm choosing not to worry about sizing here. Then mount a 'ramdisk' to this these locations, e.g.Īnd then again for your cache and localcache dirs. ![]() As we are only talking about Temp and Cache data we can just put this all into memory rather than writing to disk at all (this has the advantage of freeing up some IO for things that do actually need it too)įind out where you temp and cache dirs are (usually under moodledata, but also check your config.php to see if they have been manually set elsewhere). As you only have one web server you don't need to worry about sharing these folders between nodes so ignore any discussion about that. The short solution for you is to point moodles temp and cache directories to a faster storage medium. You have an issue with the write speed to some of your data locations, and specifically this test is showing results for tempdir. Should I take some ram and turn it into a RAMdisk for moodledata/temp? I'm at a loss to how to improve this though. The CPU's aren't the newest or fastest and I know that. Authentication isn't something I'm super concerned with because we authenticate through Google anyway. The write speed is what is killing me here. Last year I only had 1 physical CPU and everything was on the same VM, but I've since added a second CPU, added the 12,000 RPM SAS drive, and doubled the ram, but it "feels" "slower". I currently have a RAID 1 setup for both drives. Hard drives - webserver is on a SAS 7200 rpm enterprise server drive and the SQL server is on a 12000RPM SAS drive. The server has 2 Intel E-5 2620v2 chips - each has 6 cores, + hyperthreading 2.1ghzĮach VM has roughly half of the available resources (i.e. Because of CPU core limitations in the version of VMware we are using I can't utilize all my CPU cores with just one VM. I'm running VMware with 2 VM's - 1 for the web server, 1 for the database server. I ran this test on my production server and had some disturbing results.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |