GIANT_CRAB
New Member
MySQL
wlanboy
Content Contributer
I need about 512 MB of RAM to run MongoDB and about 1 GB of RAM for Tomcat.
Both MongoDB and Tomcat are running fine on OpenVZ. But I still prefer KVM with more than 1/2 GB of RAM for a MongoDB server.
So 2 GB of RAM can easily be touched by a vps.
Everything beyond that (2 GB) e.g. Glassfish, JBoss, Oracle needs a dedicated machine.
Both MongoDB and Tomcat are running fine on OpenVZ. But I still prefer KVM with more than 1/2 GB of RAM for a MongoDB server.
So 2 GB of RAM can easily be touched by a vps.
Everything beyond that (2 GB) e.g. Glassfish, JBoss, Oracle needs a dedicated machine.
wlanboy
Content Contributer
True. 500 concurrent users doing comments and your hoster will get a ticket like "may I have some additional RAM again?"
jcaleb
New Member
i host some grails apps i write for clients, so they can see it without meeting face to face. 1 grails app on tomcat + mysql around 700mb on kvm, the 300mb free is leg room for other things.
if i put this into production on small number of users, i think i need at least 2gb ram
if i put this into production on small number of users, i think i need at least 2gb ram
drmike
100% Tier-1 Gogent
I wish I could shoe horn everything into small RAM.
Not a VPS, but my big projects suffer from: Java, MySQL and memcached. I can step down the values some. Currently parked at 8GB+ steady-ish RAM consumption. Reducing MySQL cache values would drop it maybe 2GB+. memcached, can shave that too by 2GB+. Downsized is more slow requests to MySQL for data from disk.
Not a VPS, but my big projects suffer from: Java, MySQL and memcached. I can step down the values some. Currently parked at 8GB+ steady-ish RAM consumption. Reducing MySQL cache values would drop it maybe 2GB+. memcached, can shave that too by 2GB+. Downsized is more slow requests to MySQL for data from disk.
1. Pretty much anything written in java or jvm languages (clojure, scala) needs a lot of ram.
2. Since (at good hosts) there are fewer high-ram vps to a node than with low ram, you tend to get more cpu and i/o performance even if you don't use the ram.
Still though I've never made much use of high-ram vps's (512mb or more). The extra ram comes in handy for occasional spikes like compiles. But for ram-intensive processes I've always used dedicated servers so far. I ran a search engine for a while (few hundred GB of data, tiny compared to web search) that served queries without too high cpu load, but rebuliding the index took 100% cpu for several days. It's not really feasible to do that with vps's.
2. Since (at good hosts) there are fewer high-ram vps to a node than with low ram, you tend to get more cpu and i/o performance even if you don't use the ram.
Still though I've never made much use of high-ram vps's (512mb or more). The extra ram comes in handy for occasional spikes like compiles. But for ram-intensive processes I've always used dedicated servers so far. I ran a search engine for a while (few hundred GB of data, tiny compared to web search) that served queries without too high cpu load, but rebuliding the index took 100% cpu for several days. It's not really feasible to do that with vps's.
The CPython VM is pretty memory heavy, even when using WSGI containers like uWSGI. If you're running multiple uWSGI processes, each one can take 50-100mb RAM easily.
I wonder if pypy is any better on memory usage -- but I suspect that it is probably not. Perhaps something to look at on a rainy day...
Many of my customers run boards like vpsBoard (i.e. using invision) and this can be memory intensive because IPB likes to make lots of requests per second, and PHP wasn't exactly written with memory efficiency in mind (neither was Python for that matter).
I wonder if pypy is any better on memory usage -- but I suspect that it is probably not. Perhaps something to look at on a rainy day...
Many of my customers run boards like vpsBoard (i.e. using invision) and this can be memory intensive because IPB likes to make lots of requests per second, and PHP wasn't exactly written with memory efficiency in mind (neither was Python for that matter).
drmike
100% Tier-1 Gogent
Just a general question, but has IPB or similar employed query caching or addons for such. I cache queries like a lunatic. No way to pull performance and not have to scale into more boxes under load aside from caching.
Exactly. Most people need it for burst times. I usually count that if I am at 512 MB usage in a regular day, i should expect 1 GB at peaks or close.
Since the difference in price between 512 MB and 1 GB is small, it saves me a few eggs on my face and as such is well worth it.
I do not agree you get better cpu if you have a big ram erver on a node with many small. A lot of small threads means CPU is overworked, while a lot of BIG VPSes means there are fewer threads and the cpu spens less time switching, therefore is more available and responsive, even if the taskks themselves are much heavier.