httpd-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jeroen Geilman <jer...@adaptr.nl>
Subject Re: [users@httpd] General question - Apache server-status page
Date Mon, 13 Jun 2011 19:32:49 GMT
On 06/13/2011 09:23 PM, Rob Morin wrote:
>
> Can you please point me in some direction when you say   "Consider 
> proxying dynamic requests to a dedicated backend (which can run on the 
> same machine)."
>

1. set up apache to run the worker MPM; this has a much smaller memory 
footprint, and can start threads much faster than prefork can.
     Much, much faster.
     Scale these "middle-ends" to handle 2000+ threads simultaenously; 
this needs only a few gigs.
2. proxy all dynamic content - however identified - to a dedicated 
mod_fcgid backend that runs your PHP (and cache, etc)
     You can now control the exact resource ratio between static and 
dynamic content, and tune it to your requirements, or so your overall 
resources don't get clobbered.
3. either use the same box(es) to run the fcgid PHP backend, or delegate 
this task to separate backends, i.e. reallocate 3 of the boxes to run 
workerproxies + static content, and use the other 3 to run only dynamic 
content.

There's more than one way to skin the cat.


> Do you mean a reverse proxy to go before the webservers?
>
> A URL of sorts?
>
> Thanks for your replies so far..
>
> J
>
> Rob Morin
>
> Systems Administrator
>
> Infinity Labs Inc.
>
> (514) 387-0638 Ext: 207
>
> ilabs-email-sig
>
> *From:*Jeroen Geilman [mailto:jeroen@adaptr.nl]
> *Sent:* Monday, June 13, 2011 3:10 PM
> *To:* users@httpd.apache.org
> *Subject:* Re: [users@httpd] General question - Apache server-status page
>
> On 06/13/2011 09:01 PM, Rob Morin wrote:
>
> My problem is as it stands, the loads are too high already, at peek 
> times our Dual Quad Core Xeons with 18 gigs of ram go as high as 
> 100.00 and this is with a max clients of 600. We do use memcache and 
> eaccelerator...
>
> I mean I should have been more clear... I do know what the W and C are 
> I was just wondering why there would only W and Cs.. sorry for the 
> confusion there...
>
> When the server is at a load of 100 we still have 2 to 3 gigs of ram 
> to spare and there is never any swapping...
>
> I just do not think that the problem is server capacity, it seems to 
> be something else I think, but not sure what?
>
> In a top I see every couple of mins on all servers this...
>
> 22498 apache    15   0     0    0    0 Z  0.3  0.0   0:45.98 httpd 
> <defunct>
>
>
>
> So you're running bad scripts.
>
> Consider proxying dynamic requests to a dedicated backend (which can 
> run on the same machine).
>
> This offers much better control over resources, and will host 3 times 
> the requests easily.
>
>
>
> Right now all our server are at ...
>
> top - 14:58:31 up 7 days,  3:49,  3 users,  load average: 138.71, 
> 121.10, 85.80
>
> Tasks: 759 total,   1 running, 757 sleeping,   0 stopped,   1 zombie
>
> Cpu(s): 13.2%us,  2.5%sy,  0.0%ni, 82.9%id,  0.0%wa,  0.1%hi,  
> 1.3%si,  0.0%st
>
> Mem:  18470312k total, 12042288k used,  6428024k free,   580252k buffers
>
> Swap:  2096472k total,        0k used,  2096472k free,  8233784k cached
>
> Current Time: Monday, 13-Jun-2011 15:00:26 EDT
>
> Restart Time: Monday, 13-Jun-2011 12:54:23 EDT
>
> Parent Server Generation: 0
>
> Server uptime: 2 hours 6 minutes 3 seconds
>
> Total accesses: 1062620 - Total Traffic: 197.1 MB
>
> CPU Usage: u2966.14 s1473.45 cu0 cs0 - 58.7% CPU load
>
> 141 requests/sec - 26.7 kB/second - 194 B/request
>
> 378 requests currently being processed, 201 idle workers
>
> WW_WW_CC_WCC_.__WWWWWW_.WW__W__W___WCWWCWW_CWW_CWWCWWW_WW__C__WW
>
> W__W___.WCC____CCCWWW_WC_W_WW_C_CCWWWC__CW_W.WWWWWW_W_CW.W__WWCC
>
> WWRW_WWW_WW___WC_WW_RWW_W_W_WWW_C___WWCCC__WWWWW_W___WC_C.WWWWC_
>
> W_WWW___W__C_WWWWW_WWWWW_WW_W.WWW_WW_..___WCW___W_CWC_W_WW_WWWW.
>
> WC_WCWWCWCW_W_CC_.__.__WWC_CCCW_W_WW_W_W_WCWWWW___WWW_C__W__WWW_
>
> _CWWC_W.WCW_CWCWW_WCWCW__W_CCW._WCCW_WC__WWWWW_C_WW_WCWWW__WCCCC
>
> W_W_W_WWCWWW_WWW_WWW_CWC____..WWW_W__W_.CWC_C_WC_C_WC___WW_WWWC_
>
> _W_WW__WWW_C_WWCW__WW_W_WCW_W__CCWCCC_WWCW_..___WW_C_WWWWW_WCW_W
>
> __.__WW__W_C_CC__WCCWWWCWC_WW__W_R_WWC_W_W__WWWWWWWC_C_C_W__CRC_
>
> WWCW_C__._WW_WCWWWWWW_WW
>
> I am sad....
>
> Rob Morin
>
> Systems Administrator
>
> Infinity Labs Inc.
>
> (514) 387-0638 Ext: 207
>
> ilabs-email-sig
>
> *From:*Issac Goldstand [mailto:margol@beamartyr.net]
> *Sent:* Monday, June 13, 2011 2:36 PM
> *To:* users@httpd.apache.org <mailto:users@httpd.apache.org>
> *Subject:* Re: [users@httpd] General question - Apache server-status page
>
> That your server(s) is/are swamped.
>
> Consider adding more worker threads/processes.
>
>   Issac
>
> On 13/06/2011 21:23, Rob Morin wrote:
>
> If I see all the slots filled with W and C only, what would that 
> indicate, this is during high traffic times..
>
> We are on Centos 64 bit with Apache 2.2,  and have 6 webservers behind 
> HAproxy
>
> Thanks..
>
> Rob Morin
>
> Systems Administrator
>
> Infinity Labs Inc.
>
> (514) 387-0638 Ext: 207
>
> ilabs-email-sig
>
>
>
>
> -- 
> J.


-- 
J.


Mime
View raw message