here is some statistic: low network traffic =========================================== Json-RPC PB
average MIN 0.016 average MIN 0.011 average MAX 0.044 average MAX 0.029 AVG 0.027 AVG 0.017 90% line 0.027 90% line 0.017 CLIENTS 139 CLIENTS 142 ERRORS 0 ERRORS 0 average MIN 0.029 average MIN 0.023 average MAX 0.426 average MAX 0.159 AVG 0.151 AVG 0.062 90% line 0.124 90% line 0.052 CLIENTS 317 CLIENTS 316 ERRORS 0 ERRORS 0 average MIN 0.32 average MIN 0.115 average MAX 2.763 average MAX 1.521 AVG 1.064 AVG 0.405 90% line 0.908 90% line 0.308 CLIENTS 456 CLIENTS 479 ERRORS 0 ERRORS 0 average MIN 0.253 average MIN 0.02 average MAX 4.031 average MAX 1.486 AVG 1.123 AVG 0.296 90% line 0.859 90% line 0.189 CLIENTS 482 CLIENTS 501 ERRORS 30 ERRORS 0 hight network traffic (string 7000 length) =========================================== Json-RPC PB average MIN 0.016 average MIN 1.755 average MAX 0.114 average MAX 3.517 AVG 0.045 AVG 2.58 90% line 0.043 90% line 2.58 CLIENTS 138 CLIENTS 88 ERRORS 0 ERRORS 0 average MIN 0.333 average MIN 4.923 average MAX 1.516 average MAX 9.019 AVG 0.732 AVG 6.746 90% line 0.652 90% line 6.746 CLIENTS 288 CLIENTS 173 ERRORS 0 ERRORS 0 average MIN 0.409 average MIN 9.505 average MAX 2.372 average MAX 16.546 AVG 1.221 AVG 12.486 90% line 1.119 90% line 12.448 CLIENTS 429 CLIENTS 273 ERRORS 0 ERRORS 0 average MIN 0.414 average MIN 14.22 average MAX 3.932 average MAX 22.986 AVG 1.639 AVG 17.987 90% line 1.426 90% line 17.866 CLIENTS 496 CLIENTS 392 ERRORS 27 ERRORS 0 On Tue, Feb 16, 2010 at 10:08 PM, Phil Christensen <p...@bubblehouse.org>wrote: > On Feb 16, 2010, at 2:46 PM, Vlad Shevchenko wrote: > > I have developed two kind of servers: > > api - http server > > memory_db - dict-based server with some logic of putting data to dict > and getting it back. > > > > Api communicate with memory_db via txJSON-RPC but response time is huge > under high load. I also try implementation with replacement txJSON-RPC -> > Perspective Broker with no success. PB implementation has 10+ times faster > responses when returning data is small (dict with few keys and values like > list of 1-5 strings). However, PM much depends on size of returning value. > When I return a list of 600 items (each is string with length 20) response > time incredibly growing up. Unlike PB, json-rpc implementation keeps it > response-time near the same lever as with small return data. > > > > Can you suggest some protocol with a small response? Or should I look > into development of my own? > > It definitely depends on your use case... > > PB can still be a good solution for you if you change your architecture a > little bit. Instead of returning a 600-item list, you could return a > pb.Referencable object that can iteratively return each row at a time. > > You may also want to look into using Copyable or Cacheable PB objects. More > on that here: > > > http://twistedmatrix.com/documents/current/core/howto/pb-copyable.html > > Of course, if you need a large amount of data to be available before you > can begin calculations or whatnot, you're going to have to wait for it to be > sent no matter what protocol you use. > > -phil > > > > _______________________________________________ > Twisted-Python mailing list > Twisted-Python@twistedmatrix.com > http://twistedmatrix.com/cgi-bin/mailman/listinfo/twisted-python > -- WBR, Vlad Shevchenko
_______________________________________________ Twisted-Python mailing list Twisted-Python@twistedmatrix.com http://twistedmatrix.com/cgi-bin/mailman/listinfo/twisted-python
