Hiya,
Just starting to specify the stack for a rebuild job - looking to move onto D7 as long as the modules required for the build are sufficiently mature, otherwise it'll be D6 and probably based on Pressflow / Varnish / Memcache etc.
The site is transactional, so will require user accounts, but beyond that there won't be much (any?) user generated content bar from some contact forms / quote request forms etc.
This'll be the biggest site I've put together, so I'm interested to find some Drupal benchmarks for a site of this size, can anyone help?
What would be your architecture recommendations be for a site of this size? The client has already stated that downtime costs ~£10k per hour to the business at peak times, so it's going to have to be pretty robust. They are also concerned with speed, wanting all their pages to be quick to load (ie generation and serving to be ~ <333ms). My starter for 10 was something like:
- 2x server (3Ghz Xeon, 4GB Ram, 120GB Raided HDD)
- Databases hosted on each server, one master, one slaved
- Loadbalancer and Firewall
- Use full Drupal caching
- APC / Memcache & MySQL caching tested and tuned
- Reverse Proxy / Varnish
My main question is whether people recommend that a site of this size has a dedicated database server?
Comments
From taking a lesson from
From taking a lesson from hard drive raid configurations, RAID 5 requires at least three hard drives for better performance per price, while you generally only get high availability when using two servers (RAID 1, etc.), so as to your main question, there are recommendations that after a certain amount of ram, additional ram doesn't help as much because the os takes care of some caching by itself so in that case, going for throughput from master/slave combo is better from a high availability standpoint. If downtime costs ~£10k per hour, then you shouldn't skimp on selecting the right dedicated or cloud setup for the hardware/network portion.
As for the software side, pantheon has officially launched, which used to be based on Mercury and Amazon (congrats on the launch~). I think they offer consulting.
A thread further below in this group mentions people's thoughts on scalr, so you would also want to check out that site.
Aside from that, Pressflow 7 code has been uploaded to github and those guys offer consulting as well.
I've seen suggestions for using Percona instead of MySQL, so that could be a consideration.
Everything you've mentioned seems like what's needed, since you need proven solutions. Test cases mentioned by drupal.org shows that drupal can handle such pageviews as long as you have experienced people on your team.
Cheers,
Paul
Pantheon
I don't think Pantheon has actually announced pricing officially (though maybe that $100/month will turn out to be accurate) or launched the service yet - I think they're still in Beta. Also, I believe they ended up selecting Rackspace for their main servers as they found CPU and disk access performance to be better than alternatives such as Amazon. (They may still be using Amazon for their off-site backups though.)
Pantheon only offers a fully-hosted solution; they don't offer software you can download and install on your own servers. Though I'm sure anyone could roll their own "Pantheon" as it's more-or-less "only" a nicely-tuned project Mercury setup plus some scripts to manage environment setup, git push and db pull between three environments, and offsite backups... ;p (By which I mean it looks pretty sweet and if the price is right I'd rather let them do all that setup rather than me...)
Oh yeah, your priorities
Oh yeah, your priorities should be high availability first, followed by the best configuration for iops since you're worried about your database, then you will need to prioritize for everything else for your hardware/network needs.