Another thing their dashboard does is illustrate their infrastructure clearly.
I can’t count the number of startups I’ve worked at where there are extra services running, odd side utility boxes performing tasks, and general disorganization. In some cases engineering can’t tall you what one service or server does.
By outlining the architecture here, they create a living network diagram that everyone benefits from.
If you scroll to the very bottom of the dashboard, you have two metrics. Homepage load time, and their “questions” page. The homepage is a metric everyone can look at, as many customers will arrive at your site though this portal. The questions page will be different for everyone. But there will be some essential page or business process that it highlights.
By sifting down to just these two metrics, we focus on what’s most important. All of this computing power, all these servers & networks are all working together to bring the fastest page load times possible!
This performance page doesn’t just face the business. It also faces the customers. It lets them know how important speed is, and can underscore how serious the business takes it’s customers. Having an outage or a spike that’s slowing you down. Customers have some transparency into what’s happening.
Why does it do this? UNION is defined that way in SQL. Duplicates must be removed and this is an efficient way for the MySQL engine to remove them. Combine results, sort, remove duplicates and return the set.
Queries with UNION can be accelerated in two ways. Switch to UNION ALL or try to push ORDER BY, LIMIT and WHERE conditions inside each subquery. You’ll be glad you did!
What if we did UNION ALL? The result would look like this:
Here the WHERE clause works on this 11 record temp table:
But it would be much faster to move the WHERE inside each subquery like this:
(SELECT type, release FROM short_sleeve WHERE release >=2013)
(SELECT type, release FROM long_sleeve WHERE release >=2013);
That would be operating on a combined 3 record table. Faster to sort & remove duplicates. Smaller result sets cache better too, providing a pay forward dividend. That’s what performance optimization is all about!
Remember multi-million row sets in each part of this query will quickly illustrate the optimization. We’re using very small results to make visualizing easier.
You can also use this optimization for ORDER BY and for LIMIT conditions. By reducing the number of records returned by EACH PART of the UNION, you reduce the work that happens at the stage where they are all combined.
If you’re seeing some UNION queries in your slow query log, I suggest you try this optimization out and see if you can tweak
Join 19k others and follow Sean Hull on twitter @hullsean.
1. Slow Disk I/O – RAID 5 – Multi-tenant EBS
Disk is the grounding of all your servers, and the base of their performance. True with larger and larger main memory, much is available in cache, a server still needs to constantly read from disk and flush things from memory. So it’s a very very important component to performance and scalability.
What’s wrong with Raid 5?
Raid 5 was designed to give you more space, using fewer disks. It’s often used in a server with few slots or because ops misunderstand how bad it will impact performance. On a database server it can be particularly bad.
All writes see a performance hit. What’s worse is if you lose a disk, the RAID though technically still on line, will perform SO SLOWLY as to be offline. And a rebuild takes many hours. Worse still is the risk to lose another drive during that rebuild. What if you have order a drive and it takes a couple of days?
RAID 10 is the solution. Mirror each set of two drives, then stripe over those. Even with only four slots available, it’s worth it. Good read performance, good write performance, and protection.
What the heck is multi-tentant?
In the cloud, you share servers, network & disk just like you do apartments in a building. Hence the name. Amazon’s EBS or elastic block storage, extends this metaphor, offering you the welcome flexibility of a storage network. But your bottleneck can be fighting with other tenants on that same network.
Default servers do have this problem, however AWS has addressed this serious problem with a little known but VERY VERY useful feature called Provisioned IOPS. It’s a technical name, but means you can lock in reliable disk I/O. Just what the scalability doctor ordered.
MySQL is good at a lot of things, but it’s not ideal for managing application queues. Do you have a table like JOBS in your database, with a status column including values like “queued”, “working”, and “completed”? If so you’re probably using the database to queue work in your application.
It’s not a great use of MySQL because of locking problems that come up, as well as the search and scan to find the next task.
Luckily their are great solutions for developers. RabbitMQ is a great queuing solution, as is Amazon’s SQS solution. What’s more as external services they’re easier to scale.
Scalability becomes key to your business, as you customer base grows. But it doesn’t have to be impossible. Disk I/O, caching, queuing and searching are all key areas where you can make a big dent, in a manageable way. Juggle your technical debt too, and you’re golden!
Oracle has full text search support, why shouldn’t we assume the same in MySQL? Well MySQL *does* have this, but in many versions only with the old MyISAM storage engine. It has it’s set of corruption problems, and isn’t really very performant.
Better to use a proven search solution like Apache Solr. It is built specifically for search, includes excellent library support for developers of most modern web languages and best of all is easy to SCALE! Just add more servers on your network, or distributed globally.
For folks interested in the bleeding edge, Fulltext is coming to Innodb crash safe & transactional storage engine in 5.6. That said you’re still probably better off going with an external solution like Solr or Sphinx and the MySQL Sphinx SE plugin.
Cache, cache, and cache some more. Your webservers should use a solid memcache or other object cache between them & the database. All those little result sets will sit in resident memory, waiting for future web pages that need them.
Next use a page cache such as varnish. This sits in front of the webserver, think of it as a mini-webserver that handles very simple pages, but in a very high speed way. Like a pack of motorbikes riding down an otherwise packed freeway, they speed up your webserver to do more complex work.
Browser caching is also important. But you can’t get at your customers browsers, or can you? Well not directly, but you can instruct them what things to cache. Do that with proper expires headers. Have your system administrator configure apache to support this.
Technical debt can bite. What is it? As you’re developing an new idea, you’ll build prototypes. As those get deployed to customers, change gets harder, and past things you glossed over because problems. One team leaves, and another inherits the application, and the problems multiple. Overtime you’re building your technical debt as your team spends more time supporting old code and fixing bugs, and less time building new features. At some point a rewrite of problem code becomes necessary.