Web Hosting Talk : Web Hosting Main Forums : VPS Hosting : How to evaluate performance and determine when an upgrade is needed?

[smkied]

Hello,

I run a few different VPS servers with many sites on them. Sometimes I will put a really busy site on one but I never can really tell how its affecting the other sites and if in fact the VPS is performing sufficiently.

How can I determine if performance is sufficient? And how do I know when an upgrade is needed? I am running CentOS with Plesk 9.5.

It's difficult, because looking at bandwidth or number of visitors never really give an accurate measure of server performance... because if it's a DB heavy site for example, well then it cant handle much traffic, but if it's just HTML pages then it can handle a lot more.

So ultimately, how can I tell when I am in need of something with better performance?

[izumi777]

You can use software such as mrtg to measure the overall VPS performance.

[smkied]

MRTG? I googled it got like 20 different applications... do you have a link?

[InsDel]

All of the Google results lead to the same software - the first result is the main site.

If you want to see how a VPS is doing load-wise for the past 5/10/15 minutes, try running the command 'uptime'; if the load average remains consistently high, you probably need more CPU power, but there's many other factors.

'free -m' shows how much memory you're using out of your quota (in megabytes); the bottom-left values (after the row '+/- buffers/cache') are the ones you should be looking at.

There's also disk I/O to consider (especially on DB-heavy sites), but I'm not sure how to measure that.

The most important factor, however, is what the users think. If the users say the site loads fast, then it doesn't really make much of a difference what all the statistics tell you - they're the ones that matter.

If the site isn't fast enough, there may be ways of getting more performance out of your existing VPS. An example: if you're thrashing disk I/O but a lot of memory is still free (usually on OpenVZ), you may be able to set up caching.

[smkied]

Ok thanks for the information. I am running Plesk, so the server load as well as memory usage are both showing up for me. Also the VPS is running RAID 10 so I don't think disk IO would be the first thing to go?

Also, in regards to MRTG, I just want to verify that it's just for bandwidth usage? Because I think my hosting control panel shows me that already as well.

The problem for me is that my users are customers, not forum members or anything, so they are unlikely to tell me when the site is slow (it's not like I would contact Dell and let them know their site is slow...) so that's why I am trying to find a kind of logical way to figure out when it's time to get something with a bit more performance and more importantly also knowing which piece of the server is the one that requires the performance boost (ie. need more RAM, better IO, better CPU, etc.)

Any other suggestions? Really appreciate the advice so far. Cheers.

[smkied]

Also for the server loads.... what are acceptable numbers?

[InsDel]

MRTG is a way of graphing all kinds of statistics as monitored over time; a good example is one of Ramhost's server status pages. Do note that this is using RRDTool (which is newer) instead of MRTG.
I'm not quite sure exactly how to set it up, so if you want to set up something like this, you would be better off searching or waiting for someone more knowledgeable in that area.

As for the load averages, the short answer is 'it varies.' Some people like an average around 1, while others say a decent average is the number of cores your VPS has access to. What are your load/RAM usage numbers?

Also, I just found out that the command that can show I/O statistics is named iostat.

[foobic]

Originally Posted by smkied

Also the VPS is running RAID 10 so I don't think disk IO would be the first thing to go?

In my experience, no matter how good the disk array, I/O is nearly always the first thing to go.

vmstat is also good for monitoring i/o - eg. use "vmstat 2" to take a measurement every 2 seconds.

[SeedBit]

Use the command "top" and say what you see.

Generally the important numbers are the load averages in the top right, the %id and %wa. Also, how many CPU cores do you have access to will determine what load average is high or not.

Then use the command "free -m". Give us the number under total and the second number down under free (in the -/+ buffers/cache: row).

[EuroVPS/Director]

You my friend need sysstat (sar) - http://pagesperso-orange.fr/sebastien.godard/

A VPS owner, logs into his VPS and sees a load spike, he sees Apache processes and MySQL processes on top, in his top… His server is topped out.. He wants to know, when this started, what is causing it, is it I/O bound, is it CPU bound, how long is it going on?

With Sysstat, Link to developer all this data is provided to you quickly, and in a very easy to read format. Debian, Suse, CentOS, and of course Fedora/Redhat distributions have Sysstat in the installation media, or you can use your automated package installer to install the “sysstat” package. Anyone that manages a server, must have sysstat and must learn to use it properly.

I have a small and short article and photos about this program, and how it helps VPS owners, here.

Sar really is your best friend.

Take care

[UNIXy]

Check out this wiki page:

Know when to upgrade memory, CPU, disk, or bandwidth.

http://www.webhostingtalk.com/wiki/K...,_or_bandwidth

Regards
Joe / UNIXY