Question About LAMP Processor Choice

[l3ll10tt]

I'm thinking about building a cheap server for some LAMP development. I was trying to decide between a FX-4300 and Pentium G3450. The AMD would go in a AM3+ motherboard so I could upgrade later, and the Pentium will go in a H97 motherboard for the same reasons. All the rest of the configurations are the same.

 

[v8envy]

Wife's starting to complain about the AMD Phenom 820 in her LAMP dev box, so I've got some of the same kind of thinking going on.

I started with looking at the G3258 ($60 at microcenter) and the cheapest 1150 board I could find for a total of about $100. AMD is straight out, for single core performance even the Pentium G620 is faster than the FX-4100, the G3258 should be about 30% faster -- and that's before overclocking. I'm not sure the FX-4100 would be a perceptible improvement over the 820 she's starting to complain about.

However, after noodling it over I'm just not going to go there. Microcenter has some decent cpu + board deals, so I'm just going to go all out and splurge for a 4690K and Z97 board for her. Not because I think it'll work that much better, but because when it's invariably not enough I can say "well, at 4.5 ghz there's not much faster you can buy."

 

[jpiniero]

People actually overclock their development machines? That's just a bad idea.
I'd say i3 tbh.

 

[Cerb]

Why not just get an SSD and another 8+GB RAM for a desktop? With a good desktop, you shouldn't need a separate physical computer.

<- E3-1230V3, 16GB RAM, M500 480GB

 

[Cogman]

Why build a server when doing some cloud hosting is crazy cheap now-a-days.

For example, you can get a pretty decent server for just $5 a month at

https://www.digitalocean.com

Even the google app services is pretty darn cheap

https://cloud.google.com/products/


That being said, if you are dead set on a server I would try and balance # of cores with price. More cores is probably a better thing to code against than faster cores.

 

[v8envy]

i3 is $140, not a stretch from there to a higher clocked and unlocked i5 for $199. And yes, plenty of people overclock their dev boxes. If you don't go nuts with the OC the productivity gains from a 20% faster compile or load add up. I'm not talking about setting an OC record, I'm talking about being way within the engineering safety margins of the chip. Production server or continuous integration server? Yeah, bad idea. Dev box? No problem.

In her case she also uses that box not only as the server but also the client. SSD won't do anything but speed up load times, with any amount of RAM at all the entire server, OS, php and every file worked on will be in the RAM disk cache.

Using the existing desktop advice is sound though. I use Windows in a VM for the few times I need to run Windows on my machine, I don't see how doing the reverse (running Linux guest on a Windows host) would be bad if you have a manly enough desktop.

EDIT: php dev is a single core sort of activity. When CPU is required I saw one core at 100%, other 3 idle. 99.9% idle, but that .1% sometimes takes several seconds!

 

[Headfoot]

So is it a server, or a developing environment, or both? If it does both then RAM and IO are going to be most important, given that the bottlenecked part of process is usually the DB IO.

The best solution though is to profile your code and see what is holding you back. We don't know your application. If its basic framework-based web dev, the DB will be holding you back 90% of the time you're bottlenecked. Processor is going to be pretty irrelevant between those two choices, get whatever is cheaper and allows a good amount of RAM (16gb probably)

 

[v8envy]

That's a very good point, how big is your typical development data set? For most PHP dev I've seen the databases are quite small, on the order of a couple hundred megs. If that's true than RAM performance and CPU performance will be your bottlenecks (all the disk pages for your whole DB will be cached in RAM, Linux uses all unused RAM as a read-through disk cache). Relational ops with all the data in RAM are sorts and copies, which is limited only by CPU and RAM bandwidth.

For my wife's app the data set is several gigabytes, but everything still fits into the 15 or so gigs of free RAM. Relational ops, whether indexed or table scan backed are what soaks up all available single core cycles for lengthy operations. PHP and Apache themselves use next to no resources compared to that, and linux (the kernel) runs on some amazingly feeble hardware.

If your app isn't doing very much DB work then you'll be good with something as low oomph as a raspberry pi as your server. http://www.penguintutor.com/linux/raspberrypi-webserver is a thing.

So, your answer is: fire up a VM with 4 to 8 virtual cores and varying amounts of RAM. Run your app. Do some dev work and use the plethora of monitoring tools to determine whether your bottleneck is RAM, I/O or CPU (and how many). Based on that either buy whatever is cheapest (pi or cloud included) or decide where you need to spend the $: start I/Os (SSD), memory or CPU.

 

[Headfoot]

Yeah, the bottleneck can shift very quickly depending on whether your dataset fits in memory, if its mostly reads, mostly writes, etc...

 

[l3ll10tt]

I'm just trying to learn PHP & java script, possibly some ruby. So I wanted a development server that I could learn how to build applications & things.

 

[jhu]

I'm just trying to learn PHP & java script, possibly some ruby. So I wanted a development server that I could learn how to build applications & things.

In that case, can't go wrong with either. You could also save your money and run all that on your current computer and have that do both client and server duties.

 

[Headfoot]

If you're just learning, nearly anything will do.

In fact, I advocate that you intentionally use the slower server. When you have a huge hardware budget it allows you to get sloppy in optimization. When I started learning I wrote all my code on a 10 inch netbook running the original atom and an old pentium 4 laptop for the server. I was learning web development like yourself. Ruby, perl, python, some .net.

Running on slower stuff allows you to see when the program is slowing down a little better, imo. You'll notice when a CPU heavy routine causes a particular page to load slower than the others and you can address it, as opposed to waiting until it gets hit with a number of users and then the performance issues creep up after you've put it out somewhere less accessible than your development environment.

 

[l3ll10tt]

Thanks everyone for your feedback!!!