I need to design a system that can handle 200K request per second in each machine over HTTP.
The wcf service need to be hosted under win service.
I wonder if wcf can handle such a requirement?
What is the best system setup/ best configuration?
The machine itself is pretty heavy 32G RAM and 8 core (or more), and can be upgraded if needed
Can I handle such amount of request in each single machine with wcf using http?
Doing this on a single machine is likely to be pretty tough (if indeed it's possible). It would be better to make your system scale horizontally, so you can add lots of machines as required. How you do that will depend on what your system actually needs to do. If it's some simple calculation which requires no persisted state, it shouldn't be too hard. If you've got some interaction with storage of some form which really needs to be read/written on each request, it'll be a lot harder - and choosing your persistence technology is likely to be pretty key to making it all hang together.
Note that there are other benefits to scaling horizontally too - in particular, the ability to upgrade the system without any downtime (if you're careful) and removing a huge single point of failure.
You need to give some more info on this.
Do you get the request and have to process it immediately?
Can you store the request data and delegate the processing to some other thread/process? Is there any way to scale the system out instead of up?
Is this in fact the only piece of infrastructure you can deploy stuff to?
I would start by asking what is it that I want to do during request handling. then what the bottlenecks are going to be.
Related
I was reading up on problems with server based authentication. I need help with elaboration on the following point.
Scalability: Since sessions are stored in memory, this provides problems with scalability. As our cloud providers start replicating servers to handle application load, having vital information in session memory will limit our ability to scale.
I don't seem to understand why "... having vital information in session memory will limit our ability to scale", will limit the ability to scale. Is it just because the information is being replicated.. so it's to do with redundancy? I don't think so. Anyway, would anyone be kind enough to explain this further? Much appreciated.
What's being referred to is the difference between stateless and stateful server-side ops. Stateful servers keep part of their resources (main memory, mostly) occupied for retaining state pertaining to some client, even when the server is actually doing nothing at all for the client and just waiting for the client to come back. Such systems' performance profile is "linear" only up to the point where all available memory has been filled with state, and beyond that point the server seems to essentially stall. Stateless servers only keep resources occupied when they're actually doing something, and once finished doing stuff, those resources are immediately freed and available for other clients. Such servers are essentially not capped by memory limits and therefore "scale easier".
Also, the explanation given seems to refer to scenario's where a set of distinct machines present themselves to the outside world as being one, when actually they are not (this is often called a "cluster" of machines/servers). In such scenario's, if a client has connected to the "big single virtual machine", then actually he is connected to just one of the "actual machines" in the cluster. If state is kept there, subsequent visits by that same client must then be routed to the same physical machine, or that piece of state must be trafficked around to whatever machine the next visit happens to be to. The former implies the implementation of management functions that take their own set of resources, plus limitations on the freedom the cluster has to distribute the load (the opposite of why you want to do clustering), the latter implies additional network traffic that will cap scalability in essentially the same way as available memory does.
Server-based authentication makes use of sessions, which in turn make use of a local session id. In the cloud, when the servers are replicated to handle application load, it becomes difficult for one server to know which sessions are active on other servers. Now to overcome this problem, extra steps must be performed... for instance to persist the session id on to the database. However, as the servers are increasingly replicated, it becomes more and more difficult to handle all this. Therefore, server-based or session-based authentication can be problematic for scalability.
I was assigned to update existing system of gathering data coming from points of sale and inserting it into central database. The one that is working now is based on FTP/SFTP transmission, where the information is sent once a day, usually at night. Unfortunately, because of unstable connection links (low quality 2G/3G modems), some of the files appear to be broken. With just a few shops connected that way everything was working smooth, but along with increasing number of shops, errors became more often. What is worse, the time needed to insert data into central database is taking up to 12 - 14h (including waiting for the data to be downloaded from all of the shops) and that cannot happen during the working day as it would block the process of creating sale reports and other activities with the database - so we are really tight with processing time here.
The idea my manager suggested is to send the data continuously, during the day. Data packages would be significantly smaller, so their transmission and insertion would be much faster, central server would contain actual (almost real time) data and night could be used for long running database activities like creating backups, rebuilding indexes etc.
After going through many websites, I found that:
using ASMX web service is now obsolete and WCF should be used instead
WCF with MSMQ or System Messaging could be used to safely transmit data, where I don't have to care that much about acknowledging delivery of data, consistency, nodes going offline etc.
according to http://blogs.msdn.com/b/motleyqueue/archive/2007/09/22/system-messaging-versus-wcf-queuing.aspx WCF queuing is better
there are also other technologies for implementing message queue, like RabbitMQ, ZeroMQ etc.
And that is where I become confused. With so many options, do you have any pros and cons of these technologies?
We were using .NET with Windows Forms and SQL Server, but if it would be necessary, we could change to something more suitable. I am also a bit afraid of server efficiency. After some calculations, server would be receiving about 15 packages of data per second (peak). Is it much? I know there are many websites without serious server infrastructure, that handle hundreds of visitors online and still run smooth, but the website mainly uploads data to the client, and here we would download it from the client.
I also found somewhat similar SO question: Middleware to build data-gathering and monitoring for a distributed system
where DDS was mentioned. What do you think about introducing some middleware servers that would cope with low quality links to points of sale, so the main server would not be clogged with 1KB/s transmission?
I'd be grateful with all your help. Thank you in advance!
Rabbitmq can easily cope with thousands of 1kb messages per second.
As your use case is not about processing real time data, I'd say you should combine few messages and send them as a batch. That would be good enough in order to spread load over the day.
As the motivation here is not to process the data in real time, then any transport layer would do the job. Even ftp/sftp. As rabbitmq will work fine here, it's not the typical use case for it.
As you mentioned that one of your concerns is slow/unreliable network, I'd suggest to compress the files before sending them, and on the receiving end, immediately verify their integrity. Rsync or similar will probably do great job in doing that.
From what I understand, you have basically two problems:
Potential for loss/corruption of call data
Database write performance
The potential for loss/corruption of call data is being caused by a lack of reliability in the transmission of data from client to service.
And it's not clear what is causing the database contention/performance issues, beyond a vague reference to high volumes, so this answer will be more geared towards solving the first problem.
You have correctly identified the need for reliable asynchronous communication transport as a way to address the reliability issues in your current setup.
Looking at MSMQ to deliver this is a valid first step. MSMQ provides reliable communication via a store and forward messaging semantic which comes out of the box and requires very little in the way of configuration.
Unfortunately, while suitable for your needs, MSMQ relies on 2 things:
A reliable network protocol, and
A client service running on both sending and receiving machine.
From your description above, I don't believe 1 exists (the internet is not a reliable network), and you might well struggle with 2 - MSMQ only ships with Windows Server or business/enterprise versions of Windows on the desktop.(*see below...)
As a possible solution to the network reliability problem, you could use a WCF or a RESTful endpoint (using Nancy or WebApi) to expose a service operation(s) exposed over HTTP, which would accept the incoming calls from the client machines. These technologies are quite different, so you'll need to make sure you're making the correct choice early on.
WCF supports WS-ReliableMessaging from the SOAP 1.2 specification out of the box, which allows for reliable web service calls over http, however it's very config-heavy and not generally a nice framework to work with.
REST much simpler than WCF in .Net, is very lightweight and easy to use. However, for reliable delivery you would have to expose some kind of GET operation (in addition to a POST to allow the client to send data) to be called (within a reasonable time-frame) to verify the data was committed. The client would have to implement some kind of retry semantic if the result of the GET "acknowledgement" was negative.
Despite requiring two operations rather than one for the WCF route, I would favour the REST approach. I've done plenty of both and find REST services way nicer to work with.
(*) That's not to say that MSMQ wouldn't work in your ultimate solution, just that it would not be used to address the transmission reliability issue. However it could still be used to address another of your problems, that of database write contention. If you were to queue incoming requests once they came into the server, then these could be processed by an "offline" process, which could then perform the required database operations in a reliable manner. This could be done by using MSMQ transactional queues.
In response to comments:
99% messages are passed from shop to main server, but if some change
is needed (price correction, discounts etc.), that data has to be sent
to shop.
This kind of changes things. Had I understood from the beginning that you had a bidirectional requirement, and seeing as how you have managed to establish msmq communication, I would have nudged you towards NServiceBus, which is a really, really cool wrapper around MSMQ. The reason I would have done this is that you appear to have both a one way, and a publish-subscribe requirement, which is supported really nicely by NServiceBus.
We are working with a .NET 3.5 app which is fast approaching legacy status. We have an existing SOAP service which reads records from our database and saves them to a third party MS SQL database, sending all the data rows in a single batch.
This has always worked fine, but recently we've taken on a much larger client than any we've had before, and they are transmitting much larger batches, so much so that they have begun to fail. We've upped the time out and max memory sizes in IIS, and maxed out the maxRequestLength in the web.config, but we are still bumping up against size problems.
So, I understand that long term, we should consider moving away from SOAP and into WCF, and plans for that are in the works. But in the mean time, we need a short term fix for this new client. And of course, to make the business and sales people happy, we need it kinda quickly.
I'm wondering what the best-practice approach might be. Initially I'm thinking something like this, but I could be thinking inside the box too much:
Establish a bench mark of # of records over which we don’t want to attempt to sync all at once.
Before attempting to save the data, check the number of records against that bench mark
If it's above it, then break the transmission down into segments which are each below that benchmark. SELECT TOP 10000 * FROM table WHERE sent = false, etc., if the benchmark is 10000. Then update sent to true for those records once submitted. Repeat.
Obviously, this will slow the process down, so to handle the user experience, we may want to toss in a status bar so they can see the progress.
Am I on the right track?
In addition to the comments from John, you should consider if you are solving the problem in the most optimal way.
It looks like you are triggering a one way sync between 2 database by calling a web service. This approach leads to the time out and memory problems that you are experiencing.
If your goal is to do the one way sync, you could use a free framework such as Microsofts sync framework: http://msdn.microsoft.com/en-US/sync
Ok, I have a pretty complex silverlight app that gets its data from a WCF service (asp.net hosted service layer) which in turn calls into a data layer that calls stored procedures in a SQL 2005 DB to extract the needed data. So the round trip goes like this:
Silverlight App --> WCF Service --> Data Layer --> DB --> Data Layer --> WCF Service transforms Data Entity into corresponding DTO (Data Transfer Object) or List<> thereof --> Silverlight App
Much of the data is highly relational (so it needs to exist in the DB), but it will change infrequently. It seems that I have several choices of locations to cache this "semi-constant" data:
I can cache it in the data layer. My data layer is already set up to use the SQLDependency class and cache the results from a stored procedure call. I think that this is or can be an application level cache.
I can cache the resulting DTO in an application level (or session level depending on the call) cache within the WCF service itself.
2(a) I could even take this a step further by serializing the XML for the resulting DTO(s) into a file on the WCF service side so that I could (a) check memory cache, then (b) check file cache and (c) hit the data layer
I could do something similar to 2(a) with isolated storage on the client side within the SL app. I could serialize the data to the local isolated storage with a hash (or a moddate or something) and then just make a call to check that.
One more thing to add: I am hosting this WCF service in IIS7 with dynamic compression turned on so that the (often very large and easily compressed) XML response gets gzip-ed. Ideally, it would seem, I would like IIS to cache this gzip-ed result to avoid all the extra processing. I think that it may do this already but I am not sure.
I am pretty sure that the final answer to this is some flavor of "it depends", but I would love to hear how others are approaching this. A good tactical recipe of Do X, Test Performance with tool Y, the do Z if needed would be great to have.
A few links (I will add to this as I research this):
WCF Caching Approach
If you have data that are user that will change quite rarely and need fast response, going for a custom mechanism bases on local storage is a great advantage quite faster than having to wait for a server roundtrip.
Dino Sposito published an interesting article about local storage and caching on MSDN Magazine there you can find as well an approach to catch assemblies (imagine just loading the minimum package required and just go loadin the rest of assemblies in background, ... performance rocket, more complexity on your code :)).
As you said is matter to go putting in a balance and decide.
HTH
Braulio
My approach would be this:
Determine if there is actually a problem with performance (isn't it alreade acceptable to my users?)
Measure the performance at each teir (how long does it take the database to come up with data? how long does it take the service to respond with data? how much time does it take from the service to the client?)
Based on the measurements I would then determine where to do my caching. Remember that, the closer to your data storage you do caching, the easier it is, but the closer to the client you do caching, the better the performance gain (usually).
Also remember that caching should not be the first thing to do to improve performance. You should also look into other performance gains as well. Are the stored procedures slow? Is there a lot of overhead in the WCF messages? Is there some inefficient processing in the service? Do I realy need all that data in one message?
HTH,
Jonathan
I think #2 is your best bet for maintainability and architecture. IIS provides caching, why not use it?
You don't want to have to reference System.Web from a data layer. Client side is not the best option either, because you'd have to write a bunch of additional code to keep the data synchronized.
Is System.Web caching even available to WCF when it's not running in ASP.NET compatible mode? Probably best not to depend on it and write your own.
On the other hand, look into Microsoft's Velocity project, which looks like it will produce a very interesting caching technology not dependant on ASP.NET.
We just recently implemented #3, the client-side caching using Isolated Storage.
In our app we have lot of drop downs and custom fields which the app used to get from the server every time it loads. Moving these data to IS really helped. The app now makes a call to check if there were any changes on the server, and if not - loads the data from the IS, otherwise ( which is pretty rare ) refreshes IS.
That eliminated a lot of WCF calls and data transfers, the SL pages' loading time is shorter, and the app in general became more scalable because of the reduced network traffic and db access.
Yes, there are some coding involved, but the benefits for the end users are essential.
Andrew
If you use RIA Services, then a simple approach is to have two separate edmx definitions. One for cached entities, one for transactional ones.
One domain context can reference the entities on another domaincontext via AddReference see.
The cached entities could be loaded immediately after user has authenticated. For simplicity, transactional data should not load until cached entities have loaded.
Depending on the size of the cache, you may also wish to consider serializing these values to local storage.
Am using Lucene API in my web portal which is going to have 1000s of concurrent users.
Our web server will call Lucene API which will be sitting on an app server.We plan to use 2 app servers for load balancing.
Given this, what should be our strategy for replicating lucene indexes on the 2nd app server?any tips please?
You could use solr, which contains built in replication. This is possibly the best and easiest solution, since it probably would take quite a lot of work to implement your own replication scheme.
That said, I'm about to do exactly that myself, for a project I'm working on. The difference is that since we're using PHP for the frontend, we've implemented lucene in a socket server that accepts queries and returns a list of db primary keys. My plan is to push changes to the server and store them in a queue, where I'll first store them into the the memory index, and then flush the memory index to disk when the load is low enough.
Still, it's a complex thing to do and I'm set on doing quite a lot of work before we have a stable final solution that's reliable enough.
From experience, Lucene should have no problem scaling to thousands of users. That said, if you're only using your second App server for load balancing and not for fail over situations, you should be fine hosting Lucene on only one of those servers and accessing it via NDS (if you have a unix environment) or shared directory (in windows environment) from the second server.
Again, this is dependent on your specific situation. If you're talking about having millions (5 or more) of documents in your index and needing your lucene index to be failoverable, you may want to look into Solr or Katta.
We are working on a similar implementation to what you are describing as a proof of concept. What we see as an end-product for us consists of three separate servers to accomplish this.
There is a "publication" server, that is responsible for generating the indices that will be used. There is a service implementation that handles the workflows used to build these indices, as well as being able to signal completion (a custom management API exposed via WCF web services).
There are two "site-facing" Lucene.NET servers. Access to the API is provided via WCF Services to the site. They sit behind a physical load balancer and will periodically "ping" the publication server to see if there is a more current set of indicies than what is currently running. If it is, it requests a lock from the publication server and updates the local indices by initiating a transfer to a local "incoming" folder. Once there, it is just a matter of suspending the searcher while the index is attached. It then releases its lock and the other server is available to do the same.
Like I said, we are only approaching the proof of concept stage with this, as a replacement for our current solution, which is a load balanced Endeca cluster. The size of the indices and the amount of time it will take to actually complete the tasks required are the larger questions that have yet to be proved out.
Just some random things that we are considering:
The downtime of a given server could be reduced if two local folders are used on each machine receiving data to achieve a "round-robin" approach.
We are looking to see if the load balancer allows programmatic access to have a node remove and add itself from the cluster. This would lessen the chance that a user experiences a hang if he/she accesses during an update.
We are looking at "request forwarding" in the event that cluster manipulation is not possible.
We looked at solr, too. While a lot of it just works out of the box, we have some bench time to explore this path as a learning exercise - learning things like Lucene.NET, improving our WF and WCF skills, and implementing ASP.NET MVC for a management front-end. Worst case scenario, we go with something like solr, but have gained experience in some skills we are looking to improve on.
I'm creating the Indices on the publishing Backend machines into the filesystem and replicate those over to the marketing.
That way every single, load & fail balanced, node has it's own index without network latency.
Only drawback is, you shouldn't try to recreate the index within the replicated folder, as you'll have the lockfile lying around at every node, blocking the indexreader until your reindex finished.