I have been running a full LAMP stack on a 128MB Leaseweb VPS for 31 days, 14 hours, 22 minutes at 99.97% uptime. Configuration:
Member
I must respectfully suggest that this configuration is admirable, yet it may not withstand significant load. I do not wish to cause offense. Let us consider that a single poorly optimized query might exhaust your available memory. I apologize if my concern is unwarranted.
I know this is a Linux crowd, but I have seen similar experiments on Windows Server 2008 R2 back in the day.
Your Alpine setup is impressive. I do wonder about real traffic, though. Have you tested with any load tools?
Your status page | 502 Bad Gateway
| traffic from thread | pipe to /dev/null
| your experiment | EOF
The graphs don't lie | but the thread visitors do
Status page is back. I had php-fpm max_children set to 5 and the thread traffic hit 23 concurrent, so yes, it fell over. Uptime is now 99.96%. I have adjusted to 8 children and added a 10 req/s limit per IP. The experiment continues.
| back | for now
| 128MB | a suggestion, not a plan
| your tuning | rearranging deck chairs
Member
I run a similar stack for a customer in Amsterdam on a 256MB Time4VPS instance. Vilnius to Amsterdam latency is fine, about 35ms. The real issue is not RAM, it is that Time4VPS storage VPS line uses HDD and the random I/O dies when MySQL does anything complex. For 128MB you need NVMe or do not bother with a database at all.
Member
Secondhand hardware perspective: I have a Dell R210 II in my basement with 8GB RAM that I got for 40 USD. It idles at 45 watts. Your 128MB VPS saves you what, 2 EUR a month? My power bill here in Detroit runs 14 cents per kWh. The math is not as obvious as people think.
Member
In Argentina that is 800 pesos at official rate, 2000 at blue. I pay 3.50 USD for a 512MB KVM at Hostinger and I sleep at night. Cheap hosting is my drug but I do not need to suffer.
Member
This is the point I attempted to make earlier. I mean no disrespect. A single machine with 128MB cannot queue properly under burst conditions. I use OpenBSD with relayd on a 512MB Vultr instance in Tokyo and I still keep a 50% memory headroom for spikes.
