PHP: tmpfs vs memcached - php

Question

I want to store information (sessions and a lot of strings) in RAM and I don't know if I should use a tmpfs or a memcached server. Someone did some benchmark and knows which one is faster? It's needed for some ajax scripts that requests informations every 1-5 seconds per user who is logged in, like a webchat in PHP. So PHP has to connect to memcache quite often.
The advantage of using tmpfs whould be that I can create a lot of files and have a structur (dirs), while I only have a key-value system in memcached, but there i could use arrays or objects to store information.
CPU load whould be interesting, too, if there is any difference.
Thanks.

Answer 1

Just a couple points
tmpfs or ramdisk are more mature than memcached (been around longer but both are stable
tmpfs is scalable (you can resize or increase as needed without loosing the contents of tmpfs
memcahced is great if you need memory on another machine or if you need to share that information between machines.
local file/socket/pipe performance is ALWAYS faster than a network socket and accessing a file in tmpfs is the same as any other file so it does not require any 3rd party libraries.
If you don't expect you data to our grow the memory on your server use tmpfs.
If you must share the data between servers or want to store more data that whats fits into the RAM on your local server use memcahed.

Answer 2

I don't really know about speed, but here are a couple of things to consider about memcached :
memcached is based on a cluster-type architecture : you can add as many physical servers as you want, install a memcached daemon on it, and you have more memory in your cluster
which means there is virtually no limit to the amount of data you can cache, using memcached : just add a couple of servers if you need more memory
on the other hand, with tmpfs, you're limited by the amount of RAM available on each server
memcached is a caching mecanism ; it's not made to store data ; which means :
when there is not enough memory left to store a new item, the old items are removed from the cache
each item has a lifetime ; when it's expired, the item is deleted from the cache
memcached is shared : you can have several PHP servers accessing a single memcached cluster
There are many existing libraries based on memcached, or created to access memcached
including some session mecanism for PHP
and several caching libraries
neither memcached nor tmpfs are made for persistence -- if you need your data to persist (i.e. still be available even after a reboot), you need to use something like a database.
In the end, not sure about tmpfs, but I would probably use memcached, at least when it comes to :
sessions
caching
Why ? Because it's :
mature -- used a lot, there are many libraries, ...
and scalable

Answer 3

Both ramdisk and memcached are blazing fast.
I don't think speed will have any importance, if you are using MySQL on your problem.
I personally would prefer Redis instead of memcached.
Here are pros / cons:
memcached may delete a key if there is no ram. Redis and files will never do so.
some software such Joomla fail to install if sessions are not in files (e.g. memcached / redis)
both memcache and redis will be able to serve several php servers, so you will not be able to use stick sessions in a cluster.
memcached is faster, then it is redis, then it is ramdisk, then is memcachedb, then is mysql, then is filesystem sessions.
ramdisk mimics normal php sessions behavior and does not need anything to be installed.
if ramdisk fails to mount, php will fallback to filesystem and still will works (assuming the server boots)
if memcached or redis stop working, the php give nasty error and not starts at all.
Hope this helps.

Answer 4

Write access to the session data must be atomic or protected by a per-session lock, else it will get corrupted. For file-based sessions it's solved by locking the file, don't know how memcached deals with it.
Using a separate ext4 partition is not that bad as You think - VFS will cache your file I/O in RAM, so You might not need anything different.
You can fine-tune ext4 (on a partition) to cache your writes, so You could get RAM-like performance and persistance for larger-than RAM sizes.
For example you get writeback cache and 60 seconds window (delayed commit) with this:
mount /dev/sda4 -t ext4 -o rw,data=writeback,nobh,commit=60
It will work fast and it won't run out-of-RAM, it will use all available RAM because the filesystem cache is dynamic.
Try it with coping say 4MB file multiple times over second file (overwriting it) you`ll get very fast writes.

Answer 5

Actually you can use memcacheDB for what you need.