It seems this is a common question/problem but despite checking out a number of proposed solutions nothing worked for us so far.
The app
It's a simple chat app, that puts a new interface on an existing app's JSON library. We proxy all the calls to their app to avoid x-domain restrictions (IE8).
ASP.net MVC3 App;
It's hosted in IIS6, W2K3 SP2. DEV svr has 1gig ram, TEST svr has 4gig ram.
The problem
When we approach 20 concurrent users, requests start lagging - no issues in Event Viewer to be found. It looks like calls are just queued. There are NO 503's returned.
What we've tried
We're using AsyncController to long-poll a 3rd party webservice for results
Hosted in IIS6
We're using the TPL to call their service in our AsyncController method
We've modified processModel and set maxWorkerThreads=100.
We've looked at this how-to but the HTTP.SYS config looks to service an infinite number of threads so we haven't bothered adding the reg keys.
The 3rd party service can handle lots of concurrent requests (and is in a web farm, so we're fairly confident we're the weakest link)
what are we missing? - any help greatly appreciated
Well... almost four weeks later and I thought I'd update this in case anyone wants to find out what helped us overcome these limitations (we're cramming around 100 simul connections on our DEV server, 1gig Xeon).
AsyncControllers
If you've got a potentially long waiting request (i.e. long polling) then use them.
Feel free to use TaskFactory but be sure to mark it as a long running process, if there is risk you could exception in your thread, be sure to use ContinueWith so you can decrement the operations, and return the error to your caller.
ServicePointManager
If you're making downstream calls (i.e. WebService/3rd party API) then make sure you have increased DefaultConnectionLimit from the default of 2 simultaneous connections.
A rough guide is 8 * Num cores so you don't starve outgoing connection resources.
See this MSDN article on DefaultConnectionLimit for more info.
IOCP vs RestSharp
I love RestSharp's API, it's fantastic but it's probably meant more for client side programming, not for proxying requests. My mistake!! Use HttpWebRequest and the Begin/End methods to make use of IOCP
If you're looking to reverse proxy or url rewrite, check out URL Rewriter, a great library available freely on CodePlex
In the end, our issue wasn't with incoming requests, it was with requests being proxied to a third party, we weren't supplying enough connections and thus they all queued up lagging the whole system. Happy to say after lots of reading, investigation and coding we've resolved it.
Related
I have a browser plugin which will be installed on 40,000 dekstops.
This plugin will connect to a backend configuration file available via https, e.g. http://somesite/config_file.js.
The plugin is configured to poll this backend resource once/day.
But there is only one backend server. So if 40,000 endpoints start polling together the server might crash.
I could think of randomize the polling frenquency from the desktop plugins. But randomization still does not gurantee that there will not be a overload at the server.
Is using websocket in this scenario solves the scalability issue?
Polling once a day is very little.
I don't see any upside for Websockets unless you switch to Push and have more notifications.
However, staggering the polling does make a lot of sense, since syncing requests for the same time is like writing a DoS attack against your own server.
Staggering doesn't necessarily have to be random and IMHO, it probably shouldn't.
You could start with a fixed time and add a second per client ID, allowing for ~86K connections in 24 hours which should be easy for any server to handle.
As a side note, 40K concurrent connections might not as hard to achieve as you imagine.
EDIT (relating to the comments)
Websockets vs. Server Sent Events:
IMHO, when pushing data (vs. polling), I would prefer Websockets over Server Sent Events (SSE).
Websockets have a few advantages, such as client side communication which allows clients to ping the server and confirm that the connection is still alive.
The Specific Use-Case:
From the description in the question and the comments it seems that you're using browser clients with a custom plugin and that the updates you wish to install daily might require the browser to be active.
This raises different questions that effect the implementation (are the client browsers open all day? do you have any control over the client browsers and their environment? can you guarantee installation while the browser is closed?).
...
IMHO, you might consider having the client plugins test for an update each morning as they load for the first time during that day (first access).
People arrive at work in different times and they open their browsers for the first time at different schedules. So the 40K requests you're expecting will be naturally scattered across that timeline (probably a 20-30 minute timespan).
This approach makes sure that the browsers and computers are actually open (making the update possible) and that the update requests are staggered over a period of time (about 33.3 requests per second, if my assumption is correct).
If you're serving a pre-written static configuration file (perhaps updated by the server daily), avoiding dynamic content and minimizing any database calls, than 33 req/sec should be very easy to manage.
Oh the joyous question of HTTP vs WebSockets is at it again, however even after quit a bit of reading on the hundreds of versus blog posts, SO questions, etc, etc.. I'm still at a complete loss as to what I should be working towards for our application. In this post I will be supplying information on application functionality, and the types of requests/responses used in our application currently.
Currently our application is a sloppy piece of work, thrown together using AngularJS and AJAX requests to a Apache server running PHP, namely XAMPP. With the launch of our application I've noticed that we're having problems with response times when the server is under any kind of load. This probably has something to do with the sloppy architecture of our server, the hardware, and the fact that our MySQL database isn't exactly optimized.
However, with such a loyal fanbase and investors seeing potential in our application and giving us a chance to roll out a 2.0 I've been studying hard into how to turn this application into a powerhouse of low latency scalability. Honestly the best option would be hire someone with experience, but unfortunately I'm a hobbyist, and a one-man-army without much experience.
After some extensive research, I've decided on writing the backend using NodeJS this time. However I'm having a hard time deciding on HTTP or Websockets. Here's the types of transactions that are done between the Server/Client.
Client sends a request to the server in JSON format. The request has a few different things.
A request id (For processing logic based on the request)
The data associated with the request ID.
The server receives the request, polls the database (if necessary) and then responds to the client in JSON format. Sometimes the server is serving files to the client. Namely images in Base64 format.
Currently the application (When being used) sends a request to the server every time an interface is changed, which on average for our application is once every few seconds. Every action on our interfaces sends another request to the server. The application also sends requests to check for notifications/messages every 8 seconds, (or two seconds depending on if they're on the messaging interface).
Currently here are the benefits I see of a stated connection over a stateless connection with our application.
If the connection is stated, I can eliminate the requests for notifications and messages, as the server can just tell the client whenever one comes available. This can eliminate x(n)/4 requests per second to the server alone.
Handling something like a disconnection from the server is as simple as attempting to reconnect, opposed to handling timeouts/errors per request, this would only be handled on the socket.
Additional security can be obtained by removing security keys for database interaction, this should prevent the possibility of Hijacking(?) of a session_key and using it to manipulate or access another users data. The session_key is only needed due to there being no state in the AJAX setup.
However, I'm someone who started learning programming through TCP game server emulation. So I understand some benefits of a STATED connection, while I don't understand the benefits of a STATELESS connection very much at all. I know they both have their benefits and quirks, but I'm curious what would be the best approach for us.
We're mainly looking for Scalability, as we had a local application launch and managed to bottleneck at nearly 10,000 users in under 48 hours. Luckily I announced this as a BETA and the users are cutting me a lot of slack after learning that I did it all on my own as a learning project. I've disabled registrations while looking into improving the application's front and backend.
IMPORTANT:
If using WebSockets, would we be able to asynchronously download pictures from the server like we can with AJAX? For example, I can make 5 requests to the server using AJAX for 5 different images, and they will all start downloading immediately, using a stated connection would I have to wait for each photo to be streamed before moving to the next request? Would this only bottle-neck a single user, or every user that is waiting on a request to be completed?
It all boils down on how your application works and how it needs to scale. I would use bare WebSockets rather than any wrapper, since it is an already easy to use API and your hands won't be tied when you need to scale out.
Here some links that will give you insight, although not concrete answers to your questions because as I said, it depends on your expectations.
Hard downsides of long polling?
WebSocket/REST: Client connections?
Websockets, and identifying unique peers[PHP]
How HTML5 Web Sockets Interact With Proxy Servers
If your question is Should I use HTTP over Websockets ?, the response is: You should not.
Even if it is faster because you don't lose time opening the connection, you lose also all the HTTP specification like verbs (GET, POST, PATCH, PUT, ...), path, body, and also response, status code. This seams simple but you'll have to re-implement all or part of these protocol things.
So you should use Ajax, as long as it is one ponctual request.
When you need to make an ajax request every 2 seconds, you need in fact that the server sends you data, not YOU request server to check Api change (if changed). So this is a sign that you should implement a websocket server.
I'm using Laravel5 and, I want to create a notification system for my (web) project. What I want to do is, notifying the user for new notifications such as;
another user starts following him,
another user writes on his wall,
another user sends him a message, etc,
(by possibly highlighting an icon on the header with a drop-down menu. The ones such as StackOverflow).
I found out the new tutorials on Laracast: Real-time Laravel with Socket.io, where a kind of similar thing is achieved by using Node, Redis and Socket.io.
If I choose using socket.io and I have 5000 users online, I assume I will have to make 5000 connections, 5000 broadcastings plus the notifications, so it will make a lot of number of requests. And I need to start for every user on login, on the master blade, is that true?
Is it a bad way of doing it? I also think same thing can be achieved with Ajax requests. Should I tend to avoid using too many continuous ajax requests?
I want to ask if Socket.io is a good way of logic for creating such system, or is it a better approach to use Ajax requests in 5 seconds instead? Or is there any alternative better way of doing it? Pusher can be an alternative, however, I think free is a better alternative in my case.
A few thoughts:
Websockets and Socket.io are two different things.
Socket.io might use Websockets and it might fall back to AJAX (among different options).
Websockets are more web friendly and resource effective, but they require work as far as coding and setup is concerned.
Also using SSL with Websockets for production is quite important for many reasons, and some browsers require that the SSL certificate be valid... So there could be a price to pay.
Websockets sometimes fail to connect even when supported by the browser (that's one reason using SSL is recommended)... So writing an AJAX fallback for legacy or connectivity issues, means that the coding of Websockets usually doesn't replace the AJAX code.
5000 users at 5 seconds is 1000 new connections and requests per second. Some apps can't handle 1000 requests per second. This shouldn't always be the case, but it is a common enough issue.
The more users you have, the close your AJAX acts like a DoS attack.
On the other hand, Websockets are persistent, no new connections - which is a big resources issue - especially considering TCP/IP's slow start feature (yes, it's a feature, not a bug).
Existing clients shouldn't experience a DoS even when new clients are refused (server design might effect this issue).
A Heroku dyno should be able to handle 5000 Websocket connections and still have room for more, while still answering regular HTTP requests.
On the other hand, I think Heroku imposes an active requests per second and/or backlog limit per dyno (~50 requests each). Meaning that if more than a certain amount of requests are waiting for a first response or for your application to accept the connection, new requests will be refused automatically.... So you have to make sure you have no more than 100 new requests at a time. For 1000 requests per second, you need your concurrency to allows for 100 simultaneous requests at 10ms per request as a minimal performance state... This might be easy on your local machine, but when network latency kicks in it's quite hard to achieve.
This means that it's quite likely that a Websocket application running on one Heroku Dyno would require a number of Dynos when using AJAX.
These are just thoughts of things you might consider when choosing your approach, no matter what gem or framework you use to achieve your approach.
Outsourcing parts of your application, such as push notifications, would require other considerations such as scalability management (what resources are you saving on?) vs. price etc'
I have a service being load tested by a third party. A few minutes after starting, we start to see requests hanging for a very long period of time and the caller ultimately times out (after 60 seconds).
They are testing with 15 users with each user using two devices at once, so a total of 30 connections.
The service is a simple façade to a more complex operation, calling an external system. Benchmarking our communications to the external system looks as though everything is responding in the time we would expect (sub 200ms).
The IIS logs reveals a bunch of very high requests (> 200sec) which ultimately do return a 200 and have Win32 error code ERROR_NETNAME_DELETD (error 64). I have checked the Service Log and can match up the response to the request (based on the SOAP message id) and can see that we do eventually respond with the correct information (although the client has long given up).
Any ideas as to what could be causing this behavior? We're hosting in IIS using wsHttpBinding and we're using WS-Security with x509 certificates (message & transport encryption).
We don't have benchmark logging inside of our service but the code is a very simple mapping of the WCF request to the server request, making the request, and mapping the response to the WCF response. We do this manually and there is no parsing involved (straight assignments).
After a detailed investigation, including getting Microsoft support involved we were hitting up against the serviceThrottling defaults, specifically the maxConcurrentSessions. We determined this from perfmon - there is a counter for this. We were unsure as to why we saw this as the service behaved when called with a .NET client.
It turns out that the Java consumer of this application, using CXF, was not respecting the WSDL (specifically the bit about WS-SecureConversation) and closing sessions out when it closed its connection.
Our solution was to jack up the maxConcurrentSessions to a high number, set the inactivityTimeout down low (to a minute) to force session abandonment. In addition, we set establishSecurityContext to false to avoid the WSS negotiation consuming an additional session.
The solution is inelegant as the service logs are littered with errors about forced session closures, but it fixed the issue we were seeing here. Unfortunately we had a requirement for WS-Security so our solution needed to stick with that.
I hope this helps someone as this was an interesting and time consuming problem to pin down.
Can somebody explain what ajax-push is? From what I understand it involves leaving HTTP connections open for a long time and reconnecting as needed. It seems to be used in chat systems a lot.
I have also heard when using ajax-push in Java it is important to use something with the NIO-connetors or grizzle serlvet api? Again, I'm just researching what it exactly.
In normal AJAX (call it pull) you ask the server for something and you get it immediately. This is fine when you want to get some data from the server now. But what if something happens on the server and the server wants to push that event to the client(s)?
Technically this is implemented using so called long polling - the browser opens the HTTP connection and waits for the response. As long as there is nothing interesting on the server side, it waits. But when something happens, the server sends the response and the client receives it immediately. This is a huge advantage over normal polling where you ask the server every few seconds - it generates a lot of traffic and still introduces noticeable latency.
The only problem with this approach is the number of pending HTTP connections. Old-school Java servlet containers aren't quite capable of handling such amount of connections due to one-thread-per-connection limitation - they quickly run out of memory. Even though the HTTP threads aren't doing anything (waiting for some other part of the system to wake them up and give them the response), they occupy memory.
However there are plenty of solutions nowadays:
Tomcat NIO connectors
Atmosphere Ajax Push/Comet library
Servlet 3.0 #Async (most portable)
Container-specific features, but Servlet 3.0, if available, should be considered superior.