Dear All, Thank you very much for all the information you provided me with.
I guess APR might become a hassle in the long run, so we will go with standalone Tomcat. Regarding the 18,000hits/hour peaktime, obviously we cannot justify the clause: "18,000hits/hour / (60*60) = 5hits/s" Therefore we must assume peaks that are concentrated into much higher values (per second). Since static content are minimal and a master class + JSP combination is handling the assembling of general dynamic html page content, software design is on safe ground. We are also utilizing our own proprietary connection pool manager class and its abstractions so that: tomcat opens the maximum number of connections (we decide for the number - it is going to be the same number with tomcat's configuration and mysql's configuration) to the database and never closes them. Connections are always in the pool, yet either active or waiting, but always open. So we overcome the overhead of initializing connections. We have throughly tested this on a multithreaded environment and results are fantastic. So all this said, can anyone suggest values for (regarding the hardware and o/s details I provided): 1) how much RAM memory should be allocated (both initial and peak values) for the JVM 2) tomcat's connections, threads etc values (server.xml configuration) - a sample server.xml file would be great. (Tomcat's out-of-the-box server.xml is somewhat exhaustive to follow through) One last question: As I previously stated, we are planning a single virtual host and approximately 1,000 aliases. Does anyone have an idea as to how many aliases can someone (not the theoretical number, but practically applicable figure would be appreciated) define in the server.xml file for a single virtual host? Kind regards, Serkan Durusoy DNA Internet Cozumleri San. ve Tic. A.S. [A] Gardenya Plaza 2 Kat:6 Atasehir - Istanbul [P] 216 456 4 156 [F] 216 456 4 157 [U] www.dna-tr.com -----Original Message----- From: Serkan Durusoy [DNA - encoding the future] [mailto:[EMAIL PROTECTED] Sent: Sunday, March 26, 2006 11:02 PM To: users@tomcat.apache.org Subject: Performance Tuning on Tomcat 5.5.16 for site with 150,000 daily hits Hello All, I am new to this mailing list, so please bare with (and warn) me if I fall into a mistake. Here is our environment followed by my question: We are currently developing a hosted content management system (CMS) solution using mainly JSP's and Java Classes. Our production environment will be: > Tomcat 5.5.16 > JDK 1.5.06 > Windows 2003 Server Web Edition (for the time being, we are not considering linux/unix - we just don't have experiences linux/unix system administrators) > MySQL 5.0.19 (MyISAM storage engine) The server machine consists of all Intel branded OEM hardware which sum up to: > Intel E7320 Chipset Server Motherboard > Dual Xeon 3.0 CPU > 4GB (4x1GB) ECC Ram > Onboard RAID 1 or RAID 0 (we haven't decided between 0 and 1 yet - but we cannot have 10 or 0+1 or any other RAID configuration) > 2 x 500GB SATAII NCQ Hard Drives Our system design implements an approach where there is only one application (hence only one document root / webapp) and only one database. Yet the number of web sites to be hosted on a single server is assumed to be around 1000. Hence, there is only one virtual host and there are 999 Alias'es defined beneath that host. Index.jsp detects the URL entered in the browser (defined on our DNS as pointed to our tomcat server) and assembles the page using static files (images, CSS etc) and html content (stored on the database) for that specific URL. We chose this approach to better utilize resources like database connection pool etc. We are assuming 1,000 web sites (aliases) which will supposedly to handle a total of 150,000 page views daily. To simplify the suggested architecture, one may think of it as a large dynamic web site with 100,000 pages (stored on DB) and 150,000 daily page views. According to surveys and internet usage statistics in our country (Turkey), our estimate of peak-time page views will be concentrated as follows: > 18,000 hits/peak-hour I really do not have any basis (or real-life experience whatsoever) to predict the number of concurrent requests (peak workload) to the server in such an example. All I can give you (except if you have any questions that I might be able to answer) is that we will have a relatively large web site with a relatively large amount of visitors. Here is my question: We have always used tomcat with its out-of-the-box configuration so we do not have any experience on advanced configuration of Tomcat, nor do we have any experience in fine-tuning. You can safely assume that our application and our database are tuned well. What are performance oriented configuration parameters on Tomcat? Where do they reside in terms of configuration files, start-up parameters etc? What would you suggest as values to those parameters in our case? I have come across with terms like max heap size (I guess this is JVM related), max-buffer-length, max-threads, max-spare-threads and so on, but I really do not know what figures to change them to. I really do not have the luxury (or means) to play around with parameters to test and see what we will get, therefore I will really appreciate the help. All I need is someone suggest me an above-average (in terms of performance) Tomcat (and possibly JVM) configuration so that I will make sure our server won't be down on its first launch day. Kind regards, Serkan Durusoy DNA Internet Cozumleri San. ve Tic. A.S. [A] Gardenya Plaza 2 Kat:6 Atasehir - Istanbul - Turkey [P] +90 216 456 4 156 [F] +90 216 456 4 157 [U] www.dna-tr.com [E] [EMAIL PROTECTED] --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
