In two minds about upgrading my webserver's RAM

Been offered a deal, not sure whether or not to take it

[trillianjedi]

My current host has offered me an upgrade from one of our servers current 2gb RAM to 4gb for $19 setup charge and an additional $49/month.

I'm not too worried about the cost, but the fact is the site runs fine and fast enough on 2gb - 1 site, PHP/MySQL based CMS. We are experiencing growth (traffic doubles about every 12 months), but I always thought when it started running a bit slow I'd offload the MySQL to a powerful box and leave this one just with Apache duties. It's a single Proc PIV 2.4ghz.

Should I do it anyway? I'm wondering whether if I don't do it now, I might not be able to do it in the future - i.e. they stop stocking this RAM type and force me to upgrade instead.

Would you take the opportunity, or just let the continuing price fall of hardware and hosting do it's thing and upgrade the whole box to a higher spec machine with more memory in 2 years time?

Any thoughts?

TJ

PS: I know a lot of you use this host and probably got the same email as me this morning (let's not mention names in accordance with TOS) - are you taking the offer or leaving it?

[kaled]

If your server is performing well enough, then it's a waste of money.

How big is the site? If most of it can be cached in the existing memory (for read operations) then adding more memory is likely to have zero effect on performance.

Depending on the hardware, adding extra memory could even reduce reliability.

Kaled.

[trillianjedi]

Thanks Kaled,

Depending on the hardware, adding extra memory could even reduce reliability.

Nice point and one I didn't even consider - thanks.

What sort of tests should I do to look at current database size etc and get the raw facts that I need? At the moment I just use the site and see it performs "fine" (pretty quick).

This is my output from TOP, if that's any use:-

CPU states: cpu user nice system irq softirq iowait idle
 total 33.3% 0.0% 3.9% 0.0% 0.9% 0.0% 61.7%
Mem: 1028296k av, 700692k used, 327604k free, 0k shrd, 30044k buff
  602164k actv, 24024k in_d, 3640k in_c
Swap: 2048276k av, 93740k used, 1954536k free  430168k cached

I'm not sure why the swapfile is in use if there's 327mb of RAM free?

Hang on:-

Mem: 1028296k av

Isn't that 1gig? I'm paying for 2!

<added>

OK, sorted the above 1gig not 2gig with support just now. It is 1gb - my mistake.

DB size is 355mb (a forum) - increasing quite quick though.

</added>

[longen]

$49 a month would pay for 1,000 word articles - which would benefit the user most? more memory or more content.

[trillianjedi]

Hi Longen,

Content isn't really the issue here, I need to work out if that extra RAM will benefit the site, and speed is the main element/reason for upgrading RAM generally.

I suppose what I'm thinking is this : spend $50/month now to cover for future growth, or are prices likely to continue to drop to a point where it's cheaper to wait and upgrade later?

TJ

[MatthewHSE]

I'm no expert in this type of stuff, but surely if the swapfile is being used, more RAM would have to make a difference . . . wouldn't it?

[kaled]

When it comes to speed, you always have to consider bottlenecks. For instance, imagine pushing the site as hard as reasonably likely. If CPU useage still averages no more than ~50%, then a faster processor won't help much. If memory usage still averages less than ~75% adding more memory won't help much. My guess would be that bandwidth is often the primary bottleneck and this might be the case on your site. However, if you have a site that uses vast databases and performs spread-sheet-like actions, increasing memory might be essential to speeding up the site.

One thing is certain - most people who sign up for this extra memory will see little or no benefit.

Kaled.

[trillianjedi]

Thanks for the insights Kaled - that's really helpful.

There is no real "bottleneck" (in terms of something that needs fixing) at the moment, the site is fast enough. I was just considering whether it's worth taking up the offer now, as it seemed like a good deal price-wise.

The problem with bargains is you can be tempted into buying stuff you don't really need ;-)

TJ

[TypicalSurfer]

You might want to throw a stress test at the server while watching top, there are a few programs that will simulate concurrent users etc., you could also roll your own little script to run at the server.

[LifeinAsia]

For $49, you could almost buy your own 2G RAM. Well, not quite, but certainly after just a few monthly payments you could.

[ronburk]

There is no real "bottleneck" (in terms of something that needs fixing) at the moment, the site is fast enough.

Maybe it is, but how would you know? Most webmasters have 0 access to real benchmarking info about their site. They just load it themselves (often from the same subnet!) and think "boy, looks plenty fast!". Local benchmarking is also deeply flawed, since bottlenecks are often only visible once you start adding in the very real, and non-trivial latency of clients who are 15 hops away.

Your web logs will not reveal whether or how many people you lose each day because they happened to arrive during a burst of traffic and got tired of waiting for a response.

[kaled]

You make a good point about it being difficult to assess website performance, however, the bottlenecks you've mentioned would not be improved by adding extra memory to the server. In fact, the only way to reliably assess the performance gain of a server modification (excluding network hardware) is with benchmarks conducted locally.

Another way to look at this is ask the question, Is it likely that I will make an additional $620 profit in a year? On some sites the answer may well be yes but I suspect they are in the minority.

Kaled.

[trillianjedi]

Thanks guys, I appreciate the input and it's been really useful.

Is it likely that I will make an additional $620 profit in a year?

No. As I mentioned, the money isn't really the issue - the site is profitable and I'm not worried about a few hundred bucks a year. The key here is I like the site to be fast, and I personally feel that affects profitability, whether or not it's measurable, directly.

In terms of bottlenecks, there are none. I know from my stats the sites peak times, and I've used it myself during those periods - it's fast enough. Actually, it's really quite quick (good SQL optimisation and Apache was configured for me by a professional).

The main point is one of growth. I think I've (you've) convinced me out of this RAM upgrade anyway. My host made me another offer to upgrade the whole box at an extremely reasonable rate and I think I'll take the opportunity to do that.

I always prefer to have something that's under-used. That's what gives scope for growth - the last thing I want is to be upgrading because I have to, in order to sort out a bottleneck. There are no bottlenecks and I want keep it that way, now and for the next 6-12 months.

TJ

[bxbase]

Mem: 1028296k av, 700692k used, 327604k free, 0k shrd, 30044k buff

That 1028296k av is whats currently active, your total is the sum of the active ,free,buffered and shared so there is 2 gigs theres

Swap: 2048276k av, 93740k used, 1954536k free 430168k cached

The swap may have only been used once but is still showing , that may be from something that was done a month ago it wont show 0 used until the server is rebooted.

if that swap is coming into use only during a rare memory intensive situation (maybe a monthly or weekly cron) than it shouldnt be a problem, it is is always in use then ram should be upgraded.

[ronburk]

the bottlenecks you've mentioned would not be improved by adding extra memory to the server.

This statement implies that you don't understand the role of TCP stack memory in keeping communications buffers from being drained. Insufficient memory absolutely can cause communications bottlenecks.

But the crux here is, you don't have a good way of knowing where your bottlenecks are at the moment. Addressing that problem will be useful to you whether you decide to add memory, upgrade, or do neither.