multithread performance problem for web service call
-
Thanks PIEBALDconsult! 1. "web service has only one process at a time" -- at client side or server side, do you mean? 2. "web service has only one process at a time" -- any documents to prove your points? I am very interested to learn from it. regards, George
1. Server side, the service doesn't run on the client. 2. If I did, I would have given a more authoritative answer.
-
Stack<Items> stack ...;
public void MyThread(object whocares){
while(go){
Item item = null;
lock(stack){
item = stack.Pop();
}
if(item != null){
//Process Item
}
}
}Then you need somewhere to run start the loop
System.Threading.ThreadPool.QueueUserWorkItem(new System.Threading.WaitCallback(MyThread));
Now just push items on the stack when they need to be called in another thread. There are many modifications to this basic theme.
Need software developed? Offering C# development all over the United States, ERL GLOBAL, Inc is the only call you will have to make.
If you don't ask questions the answers won't stand in your way.
Most of this sig is for Google, not ego.Thanks Ennis, I have some confusion about your soluton. I think the reason of why you use stack is because there are some shared object instances between the threads, correct? But in my situation, there is no shared items between my thread jobs. You can read my code again for my thread job in method "ThreadJob". Any comments? Please feel free to correct me if I am wrong. :-) regards, George
-
1. Server side, the service doesn't run on the client. 2. If I did, I would have given a more authoritative answer.
Thanks PIEBALDconsult! If the bottleneck is "web service has only one process at a time", how do you prove it or show some related documents? regards, George
-
Like so:
class Program { static Service1\[\] clients = null; public delegate void AsyncMethodCaller(int index); static void ThreadJob (int index) { // query 100 times for (int i = 0; i < 100; i++) { clients\[index\].HelloWorld(); } } static void Main(string\[\] args) { Console.WriteLine("Specify number of threads: "); int number = Int32.Parse(Console.ReadLine()); //note: this doesn't actually specify the number of threads //the number of threads will be something like "at most 15 time the number of available cores" //which means that it will only specify the number of threads that are used if //the number if small. And any number of threads may exists - but they wouldn't be used. clients = new Service1\[number\]; IAsyncResult\[\] results = new IAsyncResult\[number\]; AsyncMethodCaller TJ = new AsyncMethodCaller(ThreadJob) for (int i = 0; i < number; i++) { clients\[i\] = new Service1(); clients\[i\].EnableDecompression = true; results\[i\] = TJ.BeginInvoke(i, null, null); } DateTime begin = DateTime.Now; foreach (IAsyncResult res in results) TJ.EndInvoke(res); Console.WriteLine("Total elapsed time (s): " + (DateTime.Now - begin).TotalSeconds); Console.ReadLine(); return; } }Thanks harold, I have tried but no performance improvements. Do you have any other ideas to improve performance? I find CPU/memory/network are used a little part, so I think there is room to improve performance. regards, George
-
Thanks harold, I have tried but no performance improvements. Do you have any other ideas to improve performance? I find CPU/memory/network are used a little part, so I think there is room to improve performance. regards, George
-
Very strange.. I'm really out of ideas now.. :( There is just 1 explanation I can think of - the web server is handling all requests sequentially and non-pipelined.
Thanks harold, Two more questions, 1. My code is posted and simple. Could you reproduce my issue? I am not suspecting it is my environment issue. :-) 2. "the web server is handling all requests sequentially and non-pipelined" -- do you have any ways or documents to prove it? regards, George
-
Thanks PIEBALDconsult! If the bottleneck is "web service has only one process at a time", how do you prove it or show some related documents? regards, George
I don't know, I'm not a Web expert, I'm hoping someone with more knowledge will step in and answer that for both of us. You could possibly create a Web Service with two methods, one that sleeps for a minute and one that doesn't. Make a request to the one that sleeps, and while it's sleeping make a request to the other and see whether or not it waits for the sleeping one to complete.
-
Thanks harold, Two more questions, 1. My code is posted and simple. Could you reproduce my issue? I am not suspecting it is my environment issue. :-) 2. "the web server is handling all requests sequentially and non-pipelined" -- do you have any ways or documents to prove it? regards, George
-
I don't know, I'm not a Web expert, I'm hoping someone with more knowledge will step in and answer that for both of us. You could possibly create a Web Service with two methods, one that sleeps for a minute and one that doesn't. Make a request to the one that sleeps, and while it's sleeping make a request to the other and see whether or not it waits for the sleeping one to complete.
Thanks PIEBALDconsult, I am not sure what you are going to prove. In your scenario, you have web service (say sleep) and another web service (say sleepless). If we call on a thread from client side, first call sleep web service, then call sleepless web service, since it is synchronous call in one thread from client, the client will definitely wait for the sleep web service to complete the return results until call the second sleepless web service. Please correct me if I am wrong in understanding your points. I am confused what you are going to prove -- so obvious results. :-) regards, George
-
1. no, I would have to know what HelloWorld() does, and how it does it, and probably a lot more 2. well, if it weren't, threading would cause a speed-up. And it didn't.
Thanks harold, For 1, Here is my code at server side. Could you reproduce my problem?
/// <summary> /// Summary description for Service1 /// </summary> \[WebService(Namespace = "http://tempuri.org/")\] \[WebServiceBinding(ConformsTo = WsiProfiles.BasicProfile1\_1)\] \[System.ComponentModel.ToolboxItem(false)\] public class Service1 : System.Web.Services.WebService { \[WebMethod\] public double HelloWorld() { return new Random().NextDouble(); } }For 2, I could understand your points. But do you have any documents to prove? I think if web service could only serve requests sequentially other than simualtenaously, it is so bad and degrade performance design. I can not believe it is designed in this way. :-) regards, George
-
Thanks harold, For 1, Here is my code at server side. Could you reproduce my problem?
/// <summary> /// Summary description for Service1 /// </summary> \[WebService(Namespace = "http://tempuri.org/")\] \[WebServiceBinding(ConformsTo = WsiProfiles.BasicProfile1\_1)\] \[System.ComponentModel.ToolboxItem(false)\] public class Service1 : System.Web.Services.WebService { \[WebMethod\] public double HelloWorld() { return new Random().NextDouble(); } }For 2, I could understand your points. But do you have any documents to prove? I think if web service could only serve requests sequentially other than simualtenaously, it is so bad and degrade performance design. I can not believe it is designed in this way. :-) regards, George
- Alright it looks like it should be doing it sequentially, otherwise Random.NextDouble could fail, no method in System.Random is threadsafe. But it should barely take any time at all. 2) Documents... well. Take Amdahl's_law for example, in which P is the proportion of the program that has been made parallel. P would be zero, so you'd get the limit 1 / (1 - 0 + 0/n) where n goes to infinity. So as you can see, the answer is 1, independent of n, meaning: no benefit at all. But of course we knew that that already, since we made no improvement. 3?) I see no good reason for this to be slow, it should be blazingly fast - it's only generating a bunch of random numbers and sending them over right?
-
- Alright it looks like it should be doing it sequentially, otherwise Random.NextDouble could fail, no method in System.Random is threadsafe. But it should barely take any time at all. 2) Documents... well. Take Amdahl's_law for example, in which P is the proportion of the program that has been made parallel. P would be zero, so you'd get the limit 1 / (1 - 0 + 0/n) where n goes to infinity. So as you can see, the answer is 1, independent of n, meaning: no benefit at all. But of course we knew that that already, since we made no improvement. 3?) I see no good reason for this to be slow, it should be blazingly fast - it's only generating a bunch of random numbers and sending them over right?
Thanks harold, 1. "otherwise Random.NextDouble could fail, no method in System.Random is threadsafe." -- I agree it is not thread safe, but I disagree since it is not thread safe, web service will do sequentially. Since each time I creat a new instance of Random object. You can check my code. Any comments? 2. "Alright it looks like it should be doing it sequentially" -- how do you prove it? My current confusion is, web service has to do sequentially or could do in parallel? regards, George
-
Thanks harold, 1. "otherwise Random.NextDouble could fail, no method in System.Random is threadsafe." -- I agree it is not thread safe, but I disagree since it is not thread safe, web service will do sequentially. Since each time I creat a new instance of Random object. You can check my code. Any comments? 2. "Alright it looks like it should be doing it sequentially" -- how do you prove it? My current confusion is, web service has to do sequentially or could do in parallel? regards, George
- doh yes, sorry, I just woke up :laugh: 2) you can easily prove it informally by measuring the difference between doing some simultaneous calls vs some sequential calls - if there is no difference (or if the parallel calls took longer) than there are only 2 possible cases: 1: the simultaneous calls were handled sequentially. 2: the simultaneous calls were so short that the overhead of doing them simultaneously was as big as or bigger than the benefit. You can, of course, test which case it is. Just make the method that is called a good bit longer (say, generate 10k random numbers, then return the last one). If it benefits from threading then it was case 2, otherwise case 1.
-
- doh yes, sorry, I just woke up :laugh: 2) you can easily prove it informally by measuring the difference between doing some simultaneous calls vs some sequential calls - if there is no difference (or if the parallel calls took longer) than there are only 2 possible cases: 1: the simultaneous calls were handled sequentially. 2: the simultaneous calls were so short that the overhead of doing them simultaneously was as big as or bigger than the benefit. You can, of course, test which case it is. Just make the method that is called a good bit longer (say, generate 10k random numbers, then return the last one). If it benefits from threading then it was case 2, otherwise case 1.
Thanks harold, 1. I did the following test and I think the web service should work simultaneously. Here is my new code at server side, and I add an additional Sleep. I think if working sequentially, it will take 5 * 1000 seconds to complete all jobs (10 threads and each thread get 100 random number), but in my testing experience, it takes about 55 seconds. Does it prove web service works simultaneously? Any comments?
/// <summary> /// Summary description for Service1 /// </summary> \[WebService(Namespace = "http://tempuri.org/")\] \[WebServiceBinding(ConformsTo = WsiProfiles.BasicProfile1\_1)\] \[System.ComponentModel.ToolboxItem(false)\] public class Service1 : System.Web.Services.WebService { \[WebMethod\] public double HelloWorld() { Thread.Sleep(500); return new Random().NextDouble(); } }2. Could you reproduce my issue? I want to know whether it is my environment issue if you could not reproduce it. 3. If you could reproduce, any ideas to improve performance? regards, George
-
Thanks harold, 1. I did the following test and I think the web service should work simultaneously. Here is my new code at server side, and I add an additional Sleep. I think if working sequentially, it will take 5 * 1000 seconds to complete all jobs (10 threads and each thread get 100 random number), but in my testing experience, it takes about 55 seconds. Does it prove web service works simultaneously? Any comments?
/// <summary> /// Summary description for Service1 /// </summary> \[WebService(Namespace = "http://tempuri.org/")\] \[WebServiceBinding(ConformsTo = WsiProfiles.BasicProfile1\_1)\] \[System.ComponentModel.ToolboxItem(false)\] public class Service1 : System.Web.Services.WebService { \[WebMethod\] public double HelloWorld() { Thread.Sleep(500); return new Random().NextDouble(); } }2. Could you reproduce my issue? I want to know whether it is my environment issue if you could not reproduce it. 3. If you could reproduce, any ideas to improve performance? regards, George
- it would take 1000 * 500 miliseconds = 500 seconds if it were sequentially. Which is more than 55 seconds, so, by RAA, it is proven that it wasn't purely sequentially. Now that's a real breakthrough - it means the threading works but the task (without the sleep) is too short. 2) well, not directly. But it works like that without webservers being involved as well. 3) The threading will help when you let the web method do something useful.
-
- it would take 1000 * 500 miliseconds = 500 seconds if it were sequentially. Which is more than 55 seconds, so, by RAA, it is proven that it wasn't purely sequentially. Now that's a real breakthrough - it means the threading works but the task (without the sleep) is too short. 2) well, not directly. But it works like that without webservers being involved as well. 3) The threading will help when you let the web method do something useful.
Thanks harold, 1. "well, not directly. But it works like that without webservers being involved as well." -- confused. Especially what do you mean "without webservers being involved"? There is web server, since I run it as web service and IIS hosts it. :-) 2. "The threading will help when you let the web method do something useful." -- confused about what you mean. Could you describe something more? regards, George
-
Thanks PIEBALDconsult, I am not sure what you are going to prove. In your scenario, you have web service (say sleep) and another web service (say sleepless). If we call on a thread from client side, first call sleep web service, then call sleepless web service, since it is synchronous call in one thread from client, the client will definitely wait for the sleep web service to complete the return results until call the second sleepless web service. Please correct me if I am wrong in understanding your points. I am confused what you are going to prove -- so obvious results. :-) regards, George
Not two services and one client, one service and two clients.
-
Thanks harold, 1. "well, not directly. But it works like that without webservers being involved as well." -- confused. Especially what do you mean "without webservers being involved"? There is web server, since I run it as web service and IIS hosts it. :-) 2. "The threading will help when you let the web method do something useful." -- confused about what you mean. Could you describe something more? regards, George