Invision Community 4: SEO, prepare for v5 and dormant account notifications By Matt Monday at 02:04 PM
raindog308 Posted March 3, 2012 Posted March 3, 2012 Just curious :-) I theory, it's a 613MB OpenVZ-like VPS. In practice, I'm not sure about the CPU throttling.
altenerg Posted March 4, 2012 Posted March 4, 2012 Our entire forum is run on a t1.micro ec2 instance. Its been great going on 6 months now.
handsoffsam Posted April 13, 2012 Posted April 13, 2012 The micro instances are pretty darn low powered. Their IO capacity is terrible. The CPU can absorb spikes in usage but constant high load will kill the machine. That said, it isn't hard to move from a micro to a small, so there's no good reason not to start there. robert
raindog308 Posted April 13, 2012 Author Posted April 13, 2012 I found this article, which indicates that Micro EC2 CPU is very capped.http://gregsramblings.com/2011/02/07/amazon-ec2-micro-instance-cpu-steal/
altenerg Posted April 13, 2012 Posted April 13, 2012 t1.Micro works perfect for us! :smile: The CPU steal is known and basically only starts happening if/when the CPU is under heavy load (using 2 EC2 compute units) continuously. Our CPU only spikes for about 2 - 3 seconds during weekly web root file backups (not during nightly mysql backups) and thats about it. Even then, there's no CPU steal as the burst is very short. I've ran tests and noticed CPU steal from continuous heavy CPU load for longer duration. As you can see from the graphs below its not time to upgrade yet. 1 min avg CPU: 1 hour avg CPU: t1.Micro instances allow you to increase CPU capacity in short bursts... A small instance referred to above by poster is only 1 EC2 Compute Unit (1 virtual core with 1 EC2 Compute Unit) but a t1.Micro is up to 2 EC2 Compute Units. From my experience in managing many Ec2 powered LAMP sites for clients... if you have CPU steal issues on a t1.Micro then in most cases you might as well bypass m1.small as it will not be enough computing units and will not perform well. Instead simply stop the t1.micro instance, change its size to c1.medium or m1. medium, restart and viola! If you perform Apache bench tests on IPB you'll notice that t1.micro is faster initially (until CPU steal) than m1.small instance (the next level up). But the instance can be upgraded in less than 2 mins - including reboot - so the performance benefits of a t1.micro can be enjoyed over m1.small until you outgrow it. Of course you need to understand your load behaviors and trends and only startup IPB forums can use t1.micro until they outgrow it. I have larger sites load-balanced on multiple t1.micros and some on c1.medium.
GX93117 Posted April 18, 2012 Posted April 18, 2012 altenerg, seems you're pretty familiar with ec2 setups. Since c1.medium are 8 times the price of a t1.micro (on US East)...would it be better to run 8 micros and load balance them or just 1 c1.medium from a performance standpoint?
altenerg Posted April 18, 2012 Posted April 18, 2012 It depends a lot on how your application(s) use CPU. For IPB I don't find there to be random long/cpu heavy queries that load up PHP and/or MySQL so 4 t1.Micro instanes load balanced should be enough. The issue is you can't load test because consistent load will cause cpu stealing on all 4 instances. If this is an "established" and growing forum that you own for the long run then go with the c1.medium and pay for 1 to 3 years reserve its a lot cheaper that way... 50 - 70% cheaper in fact. Strictly from a performance only standpoint a single c1.medium would be better.
AncientMariner Posted September 20, 2012 Posted September 20, 2012 An old thread, but I'm running a board that has roughly 1,000 plus folks on it at a time with a m1.large, and it's running at an average 60% or so. The medium was just getting hammered (and SQL is on RDS). The server doesn't do anything else except serve IP.B and a CMS. I tried doing an ELB (load balancer) for it, but the boards kept getting client IP's in the 10.x.x.x range, so it made it look like no one was ever online. :smile: Any ideas on how to get the client IP passed to IP.B when behind load balancers?
altenerg Posted September 20, 2012 Posted September 20, 2012 I'd bet I could squeeze a lot more performance from that m1.large. Yea I know I have not seen the config. ;) As for load balancing on AWS I've done it using RDS in the back end as well as MySQL on one box serving as much as 4 load balanced web servers. These we not IPB setups. I did however test load balancing for a tiny forum I run... Where you'll have a problem is with the front end, sessions isn't an issues but for me it was uploading attachments and syncing them betweenall for servers or even having attachment dealt with be one server all together. Since I use cloudfront for all images its really not much (as much) saving to have them split between servers. Let me know what you choose. Or see my sig if you'd like my help. Hayden
Recommended Posts
Archived
This topic is now archived and is closed to further replies.