We are Web 2.0 company that built a hosted Content Management solution from the ground up using LAMP. In short, people log into our backend to manage their website content and then use our API to extract that content. This API gets plugged into templates that can be hosted anywhere on the interwebs.
Scaling for us has progressed as follows:
Shared hosting (1and1)
Dedicated single server hosting (Rackspace)
1 Web Server, 1 DB Server (Rackspace)
1 Backend Web Server, 1 API Web Server, 1 DB Server
Memcache, caching, caching, caching.
The question is, what's next for us? Every time one of our sites are dugg or mentioned in a popular website, our API server gets crushed with too many connections. Or every time our DB server gets overrun with queries, our Web server requests back up.
This is obviously the 'next problem' for any company like ours and I was wondering if you could point me in some directions.
I am currently attracted to the virtualization solutions (like EC2) but need some pointers on what to consider.
What/where/how to scale is dependent on what your issues are. Since you've been hit a few times, and you know it's the API server, you need to identify what's actually causing the issue.
Is it DB lookup times?
A volume of requests that the web server just can't handle even though they're shortlived?
API requests take too long to process? (independent of DB lookups, e.g., does the code take a bit to run)?
Once you identify WHAT the problem is, you should have a pretty clear picture of what you need to do. If it's just volume of requests, and it's the API server, you just need more web servers (and code changes to allow horizontal scaling) or a beefier web server. If it's API requests taking too long, you're looking at code optimizations. There's never a 1-shot fix when it comes to scalability.
The most common scaling issues have to do with slow (2-3 seconds) execution of the actual code for each request, which in turn leads to more web servers, which leads to more database interactions (for cross-server sessions, etc.) which leads to database performance issues. High performance, server independent code with memcache (I actually prefer a wrapper around memcache so the application doesn't know/care where it gets the data from, just that it gets it and the translation layer handles DB/memcache lookups as well as populating memcache).
Depends really if your bottleneck is reads or writes. Scaling writes is much harder than reads.
It also depends on how much data you have in the database.
If your database is small, but cannot cope with the read load, you can deploy enough ram that it fits in ram. If it still cannot cope, you can add read-replicas, possibly on the same box as your web servers, this will give you good read-scalability - the number of slaves from one MySQL master is quite high and will depend chiefly on the write workload.
If you need to scale writes, that's a totally different game. To do that you'll need to split your data out, either horizontally (partitioning / sharding) or vertically (functional partitioning etc) so that you can spread the workload over several write servers which do not need to do each others' work.
I'm not sure what EC2 can do for you, it essentially offers slow, high latency machines with nonpersistent discs and low IO performance on the end of a more-or-less nonexistent SLA. I guess it might be useful in your case as you can provision them relatively quickly - provided you're just using them as read-replicas and you don't have too much data (remember they have nonpersistent discs and sucky IO)
What is the level of scaling you are looking for? Is it a stop-gap solution e.g. scale vertically? If it is a more strategic scaling project, does your current architecture support scaling horizontally?
Related
I read that StackExchange uses two data centers to house all of their servers, both data centers are in the US. I'm in Ireland so I'm sure US servers are fine for me, but how can StackExchange load quickly for users in Australia if all the database servers are in the US?
I'd just like to ask, does this mean for services like MySQL, being geographically close to the server isn't as big of a deal for keeping page load times fast?
I know they use a CDN to speed up their page load time and they probably cache certain pages to speed things up, but even if I go to some really old, unpopular question I can't notice any slow-down.
The location of the database server relative to the viewer is not the significant performance factor. As a site visitor, you aren't talking to the database -- you're talking to a web application server, which is talking to the database.
Far more important, usually, is the location of the database server relative to the application server, because many applications require multiple queries and thus multiple round trips to the database in order to render a single page, and these round trips increase the time it takes for a page to be rendered. When the database is physically proximate to the application tier, that time becomes negligible.
Speaking in general web terms, in a well-managed site like SE, with all the supporting assets in a CDN, the only delay that is relevant to you is the transit time required for that one big HTTP request/response necessary to render the page content. The transit time is not negligible, because the speed of light is still finite, so round trip times to far-flung locales even on the best routes can easily be in the 200-300ms range... but if you only need to traverse it once, you still have a respectable response time.
A site that uses a lot of ajax to fetch additonal data would not fare so well with the web server so far away. If such design were needed, you'd need geographically distributed web servers, with adjacent database replicas, and geo-routing in DNS to send read-only ajax requests to the nearest web server, which could query its local replica, get a quick response, and return a quick answer.
I once moved a MySQL server -- relative to the app server -- from being ~0.5 ms away to being ~25ms away. The page load time on the site (which was already not optimal) increased from 2 sec to 10 sec. The reason? The app had been through many iterations over the years and made a lot of unnecessary requests to the database... if I remember right, even the simplest page required 13 different queries, most of which were fetching data that wasn't actually used (like fetching your score even for pages that didn't actually display your score). This inefficiency went undetected as long as the app and the db were very, very close. But, again, this was about the distance between the web server and the database, not the database and the browser.
Stack Exchange has two data centers but at last check one of them is only a hot standby/failover site. The main site does all the work under normal operations. And, SE uses MSSQL, but that, too, is immaterial, because the fundamental phenomenon at work here is a law of physics.
Perhaps StackExchange uses several copies of databases (DB Slaves) geographically distributed across different regions of the world. That explains high speed of work even with unpopular SQL-requests.
Also between Australia and West Coast of United States, direct communication via an underwater cable is possible, which ensures a high speed of operation.
How to implement dynamically updating vote count similar to quora:- Whenever a user upvotes an answer its reflected automatically for every one who is viewing that page.
I am looking for an answer that address following:
Do we have to keep polling for upvote counts for every answer, If yes
then how to manage the server load arising because of so many users
polling for upvotes.
Or to use websockits/push notifications, how scalable are these?
How to store the upvote/downvote count in databases/inmemory to support this. How do they control the number of read/writes. My backend database is mysql
The answer I am looking for may not be exactly how quora is doing it, but may be how this can be done using available opensource technologies.
It's not the back-end system details that you need to worry about but the front end. Having connection being open all the time is impractical at any real scale. Instead you want the opposite - to be able to serve and close connection from back-end as fast as you can.
Websockets is a sexy technology, but again, in real world there are issues with proxies, if you are developing something that should work on a variety of screens (desktop, tablet, mobile) it might became a concern to you. Even good-old long polls might not work through firewalls and proxies.
Here is a good news: I think
"keep polling for upvote counts for every answer"
is a totally good solution in this case. Consider the following:
your use-case does not need any real real-time updates. There is little harm to see the counter updated a bit later
for very popular topics you would like to squash multiple up-votes/down-votes into one anyway
most of the topics will see no up-vote/down-vote traffic at all for days/weeks, so keeping a connection open, waiting for an event that never comes is a waste
most of the user will never up-vote/down-vote that just came to read a topic, so your read/write ration of topics stats will be greatly skewed toward reads
network latencies varies hugely across clients, you will see horrible transfer rates for a 100B http responses, while this sluggish client is fetching his response byte-by-byte your precious server connection and what is more importantly - thread on a back end server is busy
Here is what I'd start with:
have browsers periodically poll for a new topic stat, after the main page loads
keep your MySQL, keep counters there. Every time there is an up/down vote update the DB
put Memcached in front of the DB as a write-through cache i.e. every time there is an up/down vote update cache, then update DB. Set explicit expire time for a counter there to be 10-15 minutes . Every time counter is updated expire time is prolongated automatically.
design these polling http calls to be cacheable by http proxies, set expire and ttl http headers to be 60 sec
put a reverse proxy(Varnish, nginx) in front of your front end servers, have this proxy do the caching of the said polling calls. These takes care of the second level cache and help free up backend servers threads quicker, see network latencies concern above
set-up your reverse proxy component to talk to memcached servers directly without making a call to the backend server, yes if your can do it with both Varnish and nginx.
there is no fancy schema for storing such data, it's a simple inc()/dec() operation in memcached, note that it's safe from the race condition point of view. It's also a safe atomic operation in MySQL UPDATE table SET field = field + 1 WHERE [...]
Aggressive multi level caching covers your read path: in Memcached and in all http caches along the way, note that these http poll requests will be cached on the edges as well.
To take care of the long tail of unpopular topic - make http ttl for such responses reverse proportional to popularity.
A read request will only infrequently gets to the front end server, when http cache expired and memcached does not have it either. If that is still a problem, add memecached servers and increase expire time in memcached across the board.
After you done with that you have all the reads taken care of. The only problem you might still have, depending on the scale, is high rate of writes i.e. flow of up/down votes. This is where your single MySQL instance might start showing some lags. Fear not - proceed along the old beaten path of sharding your instances, or adding a NoSQL storage just for counters.
Do not use any messaging system unless absolutely necessary or you want an excuse to play with it.
Websockets, Server Sent Events (I think that's what you meant by 'push notifications') and AJAX long polling have the same drawback - they keep underlying TCP connection open for a long time.
So the question is how many open TCP connections can a server handle.
Basically, it depends on its OS, number of file descriptors (a config parameter) and available memory (each open connection reserves a read/write buffers).
Here's more on that.
We once tested a possibility to keep 1 million websocket connections open on a single server (Windows 7 x64 with 16Gb of RAM, JVM 1.7 with 8Gb of heap, using Undertow beta to serve Web requests).
Surprisingly, the hardest part was to generate the load on the server )
It managed to hold 1M. But again the server didn't do something useful, just received requests, went through protocol upgrade and kept those connections open.
There was also some number of lost connections, for whatever reason. We didn't investigate. But in production you would also have to ping the server and handle reconnection.
Apart from that, Websockets seem like an overkill here, SSE still aren't widely adopted.
So I would go with good old AJAX polling, but optimize it as much as possible.
Works everywhere, simple to implement and tweak, no reliance on an external system (I had bad experience with that several times), possibilities for optimization.
For instance, you could group updates for all open articles in a single browser, or adjust update interval according to how popular the article is.
After all it doesn't seem like you need real-time notifications here.
sounds like you might be able to use a messaging system like Kafka, or RabbitMQ, or ActiveMQ. Your front end would sent votes to a message channel and receive them with a listener, and you could have a server side piece persist the votes to the db periodically.
You could also accomplish your task by polling your database, and by incre/decre menting a number related to a post via a stored proc... there are a bunch of options here and it depends on how much concurrency you may be facing.
My partner and I are trying to start a website hosted in cloud. It has pretty heavy ajax traffic and the backend handles money transactions so we need ACID in some of the DB tables.
Currently everything is running off a single server. Some of the AJAX traffic are cached in text files.
Question:
What's the best way to scale the database server? I thought about moving mysql to separate instances and do master-master duplication. However this seems tough and I heard I might lose ACID properties even with InnoDB? Is Amazon RDS a good solution?
The web server is relatively stateless except for some custom log files and the ajax cache files. What's a good way to scale to multiple web servers? I guess the custom log files can be moved to a reliable shared file system or DB but not sure what to do about the AJAX cache file coherency across multiple servers. (I dont care about losing /var/log/* if web server dies)
For performance it might be cheaper to go with larger instance with more cores and memory but eventually I would need redundancy so wondering what's the best way to do this cheaply.
thanks
take a look at this post. there is plenty of presentations on the net discussing scalability. few things i suggest to keep in mind:
plan early for the data sharding [even if you are not going to do it immediately]
try using mechanisms like memcached to limit number of queries sent to the database
prepare to serve static content from other domain, in the longer run - from ngin-x-alike server and later CDN
redundancy - depends on your needs. is 'read-only' mode acceptable for your site? if so - go with mysql replication + rsync of static files and in case of failover have your site work in that mode till you recover the master node. if you need high availability - then take a look either at drbd replication [at least for mysql] or setup with automated promotion of slave server to become master node.
you might find following interesting:
http://yoshinorimatsunobu.blogspot.com/2011/08/mysql-mha-support-for-multi-master.html
http://mysqlperformanceblog.com
http://highscalability.com
http://google.com - search for scalability, lamp, failover... there are tones of case studies and horror stories from the trench lines :-]
Another option is using a scaleable platform such as Amazon Web Services. You can start out with a micro instance and configure load balancing to fire up more instances as needed.
Once you determine average resource requirements you can then resize your image to larger or smaller depending on your needs.
http://aws.amazon.com
http://tuts.pinehead.tv/2011/06/26/creating-an-amazon-ec2-instance-with-linux-lamp-stack/
http://tuts.pinehead.tv/2011/09/11/how-to-use-amazon-rds-relation-database-service-to-host-mysql/
Amazon allows you to either load balance or change instance size based off demand.
I was benchmarking my production server (it's in Beta) and the results were poor to say the least. On pages without any dynamic content, 1000 Requests with a concurrency of 1 returned 73 Requests/Sec.
When I start to add MYSQL queries to the equation, things quickly spiral out of control. The same 1000 requests on my homepage produce the following results:
CPU spikes to 50%
Load spikes to 3.7 (though that doesn't always happen)
complete request:1000
failed requests:0
write errors:0
requests/sec: 2.44
transfer rate: 113.26[Kbytes/sec]
90% of requests are served within 142ms.
95% of requests are served within 3531ms (it just keeps getting worse after that).
Taking a look at top while I run the benchmark
mysqld runs as a process is consuming roughly 7% of memory and 2.5% cpu
Apache seems to spawn 7 concurrent processes at times
At other points, Apache does not show up in Top
I'm running preforked Apache on a Micro AWS instance (ubuntu) and I'll upgrade to a higher instance, but I worry that there is an underlying problem here with the code or my Apache setup.
I am deploying Django with Mod_WSGI and I set KeepAliveTimeout to 3 just in case a couple of slow processes were screwing me up.
My code for the homepage is seemingly straightforward and though it requires joins.
def index(request):
posts=Post.objects.filter(photo__isnull=False).order_by('date').distinct()[0:7]
ohouses=Open_House.objects.filter(post__photo__isnull=False).order_by('day').distinct()[0:4]
return render_to_response("index.html", {'posts':posts,'ohouses':ohouses},context_instance=RequestContext(request))
I have left the default configuration in place for MYSQL.
Could this all be attributable to running a Micro Instance? Could my instance be somewhat corrupted? Any other plausible explanations?
There's a ton that goes into quick response times. Django is pretty optimized for what it is, but relying on a framework alone will never get you where you want to be.
If you're going to use Apache, use the MPM fork, and even then disable all modules you don't absolutely need. Apache can be made to run fast, but it's not the fastest horse out there. You'll do better with something like Nginx or (cringe) Cherokee. Cherokee is a good webserver, but usability index is like zero.
Any static resources should be served directly by your webserver or better yet, off a CDN.
Assuming you've optimized your own code to not make inefficient use of queries, Django's built in, automatic query caching will help reduce the overall amount of queries needed to the database. After that, you need to employ something like memcached.
Then, there's the server itself. Depending on the size of your site, you may not need much RAM and CPU, but it's always better to have too much than not enough. It might be beneficial to put some artificial load on your server (automated testing, spidering your site, etc), and see how your system resources hold up. If you get anywhere near capping out (I'd say over 50% with simple tests like that), you need to add some more into your instance's pool.
Search online for articles on how to optimize MySQL. Out of the box, it tends to use a lot more resources than it actually needs; there's lots of room for improvement there. And, if it's not already on its own server, consider strongly offloading it to it's own server. If you're anticipating a lot of traffic, the same server responding to web requests and fetching data from a database will become a bottleneck quick.
Could this all be attributable to running a Micro Instance?
Micro instances burst to 2 CPUs for a short period of time, after which they are severely capped for several minutes. I wouldn't trust any benchmarks done on a Micro EC2 instance for that reason.
I am running a heavy traffic site and our server is beginning to get to its limits, at the moment the entire LAMP stack is on one box (not ideal).
I would like to move the database onto it's own box or onto a cloud service, but from my previous experience moving the database off the same box as the webserver increases the latency of reads quite dramatically slowing down the site.
Is using a cloud service for this going to overcome this problem, because as far as I can tell its essentially the same situation (as moving it onto a separate box in my control)? In which case why is there so much popularity around cloud based database services at the moment?
Are cloud based database services so quick that the latency of reads is so low that its almost like having it on the same box in the same datacentre?
Using a cloud service just for your database won't help your situation.
If you only move the database, you're physically placing it in a remote location - which will always increase latencies, no matter how powerful the hardware serving the content.
I would suggest that you will see a benefit in hosting your database on a separate machines from your web server, so long as they are physically next to each other sharing a dedicated network (as already suggested).
If you wanted to explore the benefits of cloud services, I would suggest only doing so if you can move both database and web server together. Furthermore, it's really only of benefit if you explore load balancing across multiple web-servers and/or replicated databases. (The ability to scale dynamically is a major benefit of cloud based platforms).
Clouds are about paying someone else to manage the infrastructure so you don't have to. They also come with some nice benefits about being able to rapidly acquire infrastructure since you don't have to wait for physical machines to be landed you can simply tap into the "cloud's" unused capacity. Sure people build features on top of this infrastructure to make it easier to scale (this is usually programming against a certain model).
If you are thinking about a cloud when are you planning on moving to 10 servers...or 100? Do you deal with traffic that comes in large bursts where the peaks in your traffic are very high?
Since you are talking about moving to a second box I don't think you need to have the cloud discussion yet. Just add a database server and use caching like e4c5 recommended.
There will be increased latency going across the network, but it shouldn't be that noticeable. Gigabit ethernet is pretty fast. When you have tried splitting the boxes, how did you access the other box? You should be using a local, internal IP address (i.e. 192.168.#.#). If you are not, then your requests may get routed over the internet, even though the boxes are physically next to each other.
Moving to a cloud won't solve your problems if the servers aren't networked properly.