We are considering using ZOPE to manage our consumer web sites. In aggregate they currently do 500,000 page views a day. We need them to scale to 2 million page views. Their not Yahoo! but it is enough of a load that I am concerned about the performance ZOPE can provide. The feature set of ZOPE looks great and but I need to work through the performance issues. I've spent quite a bit of time looking at example ZOPE sites and none appear to be huge trafficed sites. They seem to be more vertical developer sites or business-business sites that have lots of features but not necessarily tons of users banging on them at once. Does anyone have benchmarks or performance information from either a software or hardware standpoint they can send me? Chad Dyer
Chad Dyer wrote:
We are considering using ZOPE to manage our consumer web sites. In aggregate they currently do 500,000 page views a day. We need them to scale to 2 million page views.
On what hardware? OS? A mix of static and dynamic, or pure Zope? What level of graphical content? These are questions you need to have before anyone can give you an answer. That said, 2M views per day is only 2.31 per second. if each page constitutes 3-4 graphics, that is about 4-5 (call it 5) 'hits' per second, so you wind up with about 10 'hits' per second. Running ab on a Zope/Roxen server on a dual Celeron-450 w/256MB RAM, SCSI-RAID machine I can pull 1-2 times that when connecting to an SQL database, and significantly more for other types of pages. Here are the results of one run, to a 'nearly static' Zope page: ========================== Server Software: Roxen Server Hostname: www.libc.org Server Port: 80 Document Path: /index_html Document Length: 2899 bytes Concurrency Level: 10 Time taken for tests: 38.665 seconds Complete requests: 1000 Failed requests: 0 Total transferred: 3030027 bytes HTML transferred: 2901899 bytes Requests per second: 25.86 Transfer rate: 78.37 kb/s received Connnection Times (ms) min avg max Connect: 0 0 2 Processing: 212 384 627 Total: 212 384 629 ===================== (Note that this page has 2 graphics.) A run against the PTK 'recent news' method (the page itsself hase 5 graphics, 2 of which are retreived from calls to the adbanner product shows, 4 news items (ZODB stored), shows a rate of ~7 Requests per second. Now, AIUI, ab (apachebench) is counting each retreival of the specified url as a 'Request'. If so, you can take the first results and compare them to the 2.31 views/second from the 2M views/day. At that rate, the PTK request would be serving up roughly 6M views/day. Mind you, this machine has many other things running on it at the time (like several other Zope instances for example, as well as ab), so YMMV. Maybe next week I'll post ab results against an HP N-Class .... (btw, please don't send html to the list) -- In flying I have learned that carelessness and overconfidence are usually far more dangerous than deliberately accepted risks. -- Wilbur Wright in a letter to his father, September 1900
Bill, neat analysis. Chad, you'll probably be concerned with peak performance. I don't have any replicable stats on Zope performance, but we run Zope.org on a a single PIII-500 with an IDE hard disk and 256MB of RAM. It receives on average about 2 million hits a month, which translates into about 1 million page views a month (we have only three widely-viewed graphics on the site). We have stats that show averaged over a one-hour period, it can handle at least 30 page views/sec. Not burning-rubber, but respectable. Note that this is only indicative of what it *has* handled without significant impact, not what it *can* handle. Peak speed will obviously also be influenced by hardware decisions. <plug> We also sell ZEO (Zope Enterprise Option) which is a combination of service and software that allows you to spread load across multiple Zope instances. You may need it, I'm not sure. See http://www.digicool.com/Solutions/ZEOFactSheet.pdf for more info or email Paul Everitt (mailto:paul@digicool.com) </plug> Bill Anderson wrote:
Chad Dyer wrote:
We are considering using ZOPE to manage our consumer web sites. In aggregate they currently do 500,000 page views a day. We need them to scale to 2 million page views.
On what hardware? OS? A mix of static and dynamic, or pure Zope? What level of graphical content?
These are questions you need to have before anyone can give you an answer.
That said, 2M views per day is only 2.31 per second. if each page constitutes 3-4 graphics, that is about 4-5 (call it 5) 'hits' per second, so you wind up with about 10 'hits' per second. Running ab on a Zope/Roxen server on a dual Celeron-450 w/256MB RAM, SCSI-RAID machine I can pull 1-2 times that when connecting to an SQL database, and significantly more for other types of pages.
Here are the results of one run, to a 'nearly static' Zope page: ========================== Server Software: Roxen Server Hostname: www.libc.org Server Port: 80 Document Path: /index_html Document Length: 2899 bytes Concurrency Level: 10 Time taken for tests: 38.665 seconds Complete requests: 1000 Failed requests: 0 Total transferred: 3030027 bytes HTML transferred: 2901899 bytes Requests per second: 25.86 Transfer rate: 78.37 kb/s received Connnection Times (ms) min avg max Connect: 0 0 2 Processing: 212 384 627 Total: 212 384 629 ===================== (Note that this page has 2 graphics.)
A run against the PTK 'recent news' method (the page itsself hase 5 graphics, 2 of which are retreived from calls to the adbanner product shows, 4 news items (ZODB stored), shows a rate of ~7 Requests per second.
Now, AIUI, ab (apachebench) is counting each retreival of the specified url as a 'Request'. If so, you can take the first results and compare them to the 2.31 views/second from the 2M views/day. At that rate, the PTK request would be serving up roughly 6M views/day.
Mind you, this machine has many other things running on it at the time (like several other Zope instances for example, as well as ab), so YMMV.
Maybe next week I'll post ab results against an HP N-Class ....
(btw, please don't send html to the list)
-- In flying I have learned that carelessness and overconfidence are usually far more dangerous than deliberately accepted risks. -- Wilbur Wright in a letter to his father, September 1900
_______________________________________________ Zope-Dev maillist - Zope-Dev@zope.org http://lists.zope.org/mailman/listinfo/zope-dev ** No cross posts or HTML encoding! ** (Related lists - http://lists.zope.org/mailman/listinfo/zope-announce http://lists.zope.org/mailman/listinfo/zope )
-- Chris McDonough Digital Creations, Inc. Zope - http://www.zope.org
participants (3)
-
Bill Anderson -
Chad Dyer -
Chris McDonough