Private forum's index page is so slow

firegurafiku

I have a Flarum instance running on my Digital Ocean VPS in a Docker container. It's a private forum which is not supposed to be used by many users, but it still has index page (with no discussions show) visible to the Internet. Hence, it maybe a target for DDoS attack.

So I decided to make a performance test before someone else did, and I got quite sad results:

$ ab -n 100 -c 10 http://<HOST>/
...
Document Path:          /
Document Length:        7531 bytes

Concurrency Level:      10
Time taken for tests:   20.607 seconds
Complete requests:      100
Failed requests:        0
Total transferred:      788700 bytes
HTML transferred:       753100 bytes
Requests per second:    4.85 [#/sec] (mean)
Time per request:       2060.714 [ms] (mean)
Time per request:       206.071 [ms] (mean, across all concurrent requests)
Transfer rate:          37.38 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:       47   82  44.0     52     166
Processing:  1023 1925 248.9   1953    2932
Waiting:     1021 1906 234.8   1944    2930
Total:       1072 2007 266.6   2009    3080

Percentage of the requests served within a certain time (ms)
  50%   2009
  66%   2055
  75%   2090
  80%   2128
  90%   2215
  95%   2380
  98%   2617
  99%   3080
 100%   3080 (longest request)

Why would a page which shows barely nothing require two seconds too be generated?

On the contrast, when I do this test against a static file, it operates much faster:

$ ab -n 100 -c 10 http://<HOST>/assets/rev-manifest.json
...
Document Path:          /assets/rev-manifest.json
Document Length:        191 bytes

Concurrency Level:      10
Time taken for tests:   1.629 seconds
Complete requests:      100
Failed requests:        0
Total transferred:      46100 bytes
HTML transferred:       19100 bytes
Requests per second:    61.40 [#/sec] (mean)
Time per request:       162.862 [ms] (mean)
Time per request:       16.286 [ms] (mean, across all concurrent requests)
Transfer rate:          27.64 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:       47   75  37.2     51     153
Processing:    47   76  38.2     51     158
Waiting:       47   76  38.2     51     158
Total:         94  151  74.9    103     304

Percentage of the requests served within a certain time (ms)
  50%    103
  66%    195
  75%    221
  80%    244
  90%    271
  95%    301
  98%    303
  99%    304
 100%    304 (longest request)

What can I do to make things fasta?

JoshyPHP

Your cache/storage dirs may not writable, which would cause the cache to be rebuilt on every request.

firegurafiku

JoshyPHP Your cache/storage dirs may not writable,

No, they're writable.

webeindustry A static file will always be much faster. That's a no-brainer.

Surely, the hardware is slow and unpredictable, and the comparison to static file access is a bit unfair. But I wouldn't call this test a no-brainer: I know I have slow hardware, but it would be enough to serve a small number of users. What I wanted to check is whether index page for unauthorized users is cacheable (it should be, since there is no need to make a lot of DB queries to show "There are no discussions there" message). But it seems to be slow. ?

webeindustry

That's DO for you. They can't even tell you the value of a vcore because of the XEN architecture it runs under. You may or may not get a VPS with a bunch of noisy neighbors.

A static file will always be much faster. That's a no-brainer.

webeindustry

firegurafiku

It's not the hardware it's the over-committing of a hypervisor. All hosting companies do this, but some packages and platforms will be able to tell you a shared dedicated minimum of resources or have dedicated resource packages. DO can't do that, unfortunately.

I'm not a developer and can't tell you if guests are seeing a cached index page or not, but yes this will work for a small or even medium sized community. Discuss is being run on a $10 droplet, it started off on a $5 droplet! I put in some reqs/s limitations to remote addys & globally per vhost as a basic measure to guard against ddos. This is easily accomplished with the webserver.

firegurafiku

webeindustry All hosting companies do this, but some packages and platforms will be able to tell you a shared dedicated minimum of resources or have dedicated resource packages. DO can't do that, unfortunately.

Thank you for explanation. Of course, I knew DO are running their droplets on a hypervisor, but I didn't think the situation is that bad. Unfortunately, I can afford only $5/month for this site (working at a university in Russia is so uncool...), so I don't a have a lot of hosting options.

Does somebody have positive experience with Scaleway? Is it worth to take ARM cores instead of x86?

BlackSheep

firegurafiku Scaleway is great.