So far we're very happy with it, and we hope you are too! Our more technically-minded members may have a few questions…
A six-year-old Pentium, really? Closer to seven! And feature-wise, it's a decade old. This is a cache. We need disks, RAM and network bandwidth, not CPU power, and we don't want to waste your money.
The server's already handled 3x traffic bursts while serving bulk downloads - close to its 100Mbps limit - and it's pushed out almost 300GB in one day, so it should be fine.
To ensure good performance, we maxed out the RAM, reconfigured the Linux kernel to strip out features we didn't need, and recompiled it for this specific CPU. It seldom ventures above 15% usage.
As a bonus, rolling out similar changes out to our SSD-based caches in Virginia, Tokyo, Sydney and Arizona freed up ~5% of their RAM. Most of these only have 1GB to play with, so every little helps.
Why so much storage? Our members' content is growing, especially after the image and thumbnail upgrades earlier this year. Most caches only hold part of our library; but our goal is to be able to serve every public file from North America, without having to go to Europe.
Why not SSDs? We'll be storing a copy of most public content in Inkbunny's 1.5TB (and growing) library - almost 8 million files, including thumbnails. It's already soaked up 570GB, and that'll double by March. SSDs this size are not yet cost-effective to lease; spinning rust is still the way to go 'til 2019…
[Most content will be stored and read sequentially, so random access speed is less important here. Some smaller caches peak at over a hundred requests a second, but this one averages less than ten.]
Inkbunny preloads content you may need - such as the next image in a pool, or the next search page. This improves user experience over slower connections, at the cost of extra bandwidth on our end. We're fine with this - even if we'd had to pay more, the cost is low - we just need to account for it.
What about Europe? The rest of the world? Our secondary server is more than capable of handling it! Traffic has grown as members enabled wide mode/huge thumbs, but it's still well within capacity.
Why not stick with OVH? We closely considered OVH's range of offerings; but for this box, at this time, it made sense to go with our existing main provider, LeaseWeb.
Their Black Friday offers had limited availability in Quebec. Their closest SoYouStart server had only two drives - so we'd only get 2TB after mirroring them - while their Kimsufi models lack redundancy entirely. Other caches rely on this server; we want it to stay up up even if a disk drive fails.
LeaseWeb hosts our servers in the Netherlands; their Virginia datacenter is 10ms closer than Quebec. Our cache's upstream connection is also their connection home - so they'll fix any problems fast. Lastly, donors can fund our services via them, and know their money is going directly to our hosting.
While we decided not to go with OVH this time, we may use them in the future - particularly where LeaseWeb lacks VPS coverage, as in some places we do not need a dedicated server.
So you don't need any more money, right? We do have ongoing expenses totalling over $4000/year, and you're welcome to help with that. [Many have done so for several years, and we thank them!]