This post is for anyone curious about what hardware goes in to keeping Inkbunny ticking.
Inkbunny's user base is growing rapidly so it's time to upgrade our main server "Fluttershy" again!
Thanks to the generosity of our sponsors, we can easily afford to expand the server's capabilities when we need to.
We have ordered 2 x SSD disk drives to use in mirrored RAID mode for our database and we're doubling the RAM to 32GB.
The performance improvements should be huge as the web server and database will no longer be sharing a disk pack, and the expanded RAM will allow us to cache the whole database in memory.
We're expecting up to 3 hours downtime starting about 10 hours from when this journal was posted.
After the upgrade, Fluttershy's specs will be:
8 x 2.33GHz Xeon CPU cores 32GB DDR3 RAM 1TB 15kRPM SAS disk pack in RAID 5 for the web server and assets store 2 x 60GB SSD disks in RAID 1 for the database
At this stage the server load (and added cost) doesn't justify having separate database and web application servers. But that is a move we expect to make once we reach the limits of what one server can handle.
We do have a secondary server "Angel" that provides data mirroring and onsite backup, which works in conjunction with multiple offsite backup systems. But it is a low power machine designed for bulk storage only.
Maybe some time ago SSD wasn't ready for prime time. They're as reliable as any drive these days. If they fail sooner than mechanical drives then it's no big deal. We get all replacements for free.
Maybe some time ago SSD wasn't ready for prime time. They're as reliable as any drive these days. If
Unless you're spending the earth, SAS 15k is far better for database operations and typically better for RAID operations too. SSDs have an alarmingly high failure rate, you should be running them in AT LEAST RAID-10/RAID-1+spare, so 3 or 5 drives.
Hope your SSDs are SLC and not MLC. o.O Unless you're spending the earth, SAS 15k is far better for
SSDs are suuuuper fast compared to SAS. Way faster read/write and almost no seek time. An SSD failure is no big deal in Raid 1. All the hardware is replaced for free anyway. :P
SSDs are suuuuper fast compared to SAS. Way faster read/write and almost no seek time. An SSD failur
If you have a failure in RAID-1 you lose a lot of access speed. Besides, for server applications the bottleneck is never going to be disk IO unless you're hosting on a VPS, it's always going to be network latency.
Stability is far more important than raw disk IO speed and picking devices with only a 3000-5000 rewrite lifecycle probably isn't the smartest choice, imo. If they're SLC SSDs, by all means ignore the above.
If you have a failure in RAID-1 you lose a lot of access speed. Besides, for server applications the
We can get a failed disk replaced in about 3 hours. :3 Our system notifies us instantly of failures.
In our case, the network is not the major bottleneck as our network port speed is faster than most disks. Waiting for disk and CPU is the bottleneck when complex operations need to be performed for hundreds of users at once.
Judging from the specs you offered up, I'm assuming eight Xeon cores would be 2x L5410 or E5410 Xeons, which would mean you're using (FB)DDR2 RAM. There's definitely a performance gain from adding RAM, 16GB isn't very much for serving a large site.
Other performance gains you'd probably be able to attain quite easily would be: - Caching: I'm not aware if IB uses any, if it does it's quite transparent, but this can lower the disk IO usage quite significantly. - Splitting operations onto seperate VPS's: This doesn't really give you a performance increase, perse, but allows you to see where the CPU / RAM usage is in order to optimize without having to buy multiple servers. - Switching to Oracle: postgresql is not fast, though a lot of its features are better than Maria/MySQL. Oracle single connection license should be affordable with the amount of donations coming in (unless you're avoiding proprietary techs for ethical/moral reasons).
All in all, it's nice to see IB getting a much-needed upgrade. If you can get hardware replaced in 3 hours then I guess hardware with high fail rates isn't really an issue. I assume you lease, rather than buy, your hardware?
Keep up the good work. :3
Judging from the specs you offered up, I'm assuming eight Xeon cores would be 2x L5410 or E5410 Xeon