httpd-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From fernando castano <Fernando.Cast...@Sun.COM>
Subject Re: [users@httpd] apache load balance very uneven
Date Mon, 21 Jul 2008 21:35:25 GMT

André Warnier wrote:
> Hi.
>
> I know next to nothing about load balancing per se, so this will be a 
> very naive question related to the data below : does it matter ?
> I mean, I can see that the load appears to be uneven, but the grand 
> total seems to be about 10% of the available cpu time.
I am only showing an example with a few users, but I can drive the load 
up to 100% CPU utilization.  I expect the server to hold well in any 
case.  At low user level I expect the load to be even, so I get the best 
possible response.
> So do you really care if one instance is using 4% and another 0.4% if 
> there is still 90% available in total ?
>
> I am also wondering if what is shown down there is not just this 
> phenomenon : whenever the load balancer has to decide to which 
> back-end to pass a request, I suppose it checks first which ones are 
> already busy.
no, it should round-robin, no matter what is the current load in each od 
the app. servers.  That is the algorithm I am using.


> Since in this case the total load is light and most of them are always 
> free, it might just take the first of the list, which then ends up 
> more used than the others.  No ?
>
> André
>
>
> fernando castano wrote:
>> Hi all,
>>
>> I'm new to apache.  I am experiencing a problem with apache load 
>> balancer.  I configured the load balancer across 10 app servers 
>> (glassfish domains), but when I see the way the cookies (and load) 
>> are distributed I see a very uneven distribution.  Here is my proxy 
>> configuration:
>>
>>
>> fc112959@kenstgweb01$ more proxy_cluster.conf
>> # configuration for clustering more then one glassfish
>> ProxyPass / balancer://cluster/  stickysession=JSESSIONID nofailover=Off
>> ProxyPassReverse / http://kenstgapp01:8080
>> ProxyPassReverse / http://kenstgapp01:8280
>> ProxyPassReverse / http://kenstgapp01:8380
>> ProxyPassReverse / http://kenstgapp01:8480
>> ProxyPassReverse / http://kenstgapp01:8580
>> ProxyPassReverse / http://kenstgapp01:8780
>> ProxyPassReverse / http://kenstgapp01:8880
>> ProxyPassReverse / http://kenstgapp01:8980
>> ProxyPassReverse / http://kenstgapp01:9080
>> ProxyPassReverse / http://kenstgapp01:9180
>> <Proxy balancer://cluster/ >
>>       BalancerMember http://kenstgapp01:8080  route=kenstgapp01_8080 
>> loadfactor=1
>>       BalancerMember http://kenstgapp01:8280  route=kenstgapp01_8280 
>> loadfactor=1
>>       BalancerMember http://kenstgapp01:8380  route=kenstgapp01_8380 
>> loadfactor=1
>>       BalancerMember http://kenstgapp01:8480  route=kenstgapp01_8480 
>> loadfactor=1
>>       BalancerMember http://kenstgapp01:8580  route=kenstgapp01_8580 
>> loadfactor=1
>>       BalancerMember http://kenstgapp01:8780  route=kenstgapp01_8780 
>> loadfactor=1
>>       BalancerMember http://kenstgapp01:8880  route=kenstgapp01_8880 
>> loadfactor=1
>>       BalancerMember http://kenstgapp01:8980  route=kenstgapp01_8980 
>> loadfactor=1
>>       BalancerMember http://kenstgapp01:9080  route=kenstgapp01_9080 
>> loadfactor=1
>>       BalancerMember http://kenstgapp01:9180  route=kenstgapp01_9180 
>> loadfactor=1
>> </Proxy>
>> fc112959@kenstgweb01$
>>
>> And here is how the load gets distributed across jmeter 10 clients:  
>> as you can see, only 7 of the jvms get work, and among them the 
>> amount of work they do is very uneven (second to last entry in each 
>> row is % of cpu used by the process).
>> These domains are exactly the same.  I've checked the cookie 
>> distribution and reflects the load distribution (uneven).  If I 
>> increase clients I eventually get work in all jvms (still uneven), 
>> and that just proves that all jvms can be routed thru apache load 
>> balancer.  I am generating load with jmeter.  Any hints of what am I 
>> doing wrong? how to fix it?
>>
>>
>> PID USERNAME  SIZE   RSS STATE  PRI NICE      TIME  CPU 
>> PROCESS/NLWP       1388 root     3338M 3257M sleep    0    0   
>> 9:08:45 6.5% java/89
>> 1414 root     3332M 3253M cpu28    0    0   7:32:01 4.2% java/92
>> 1417 root     3333M 3253M cpu9     0    0   7:14:39 2.3% java/96
>> 1424 root     3332M 3254M cpu12    0    0   7:03:12 2.2% java/89
>> 1420 root     3332M 3254M cpu6     0    0   7:35:40 2.1% java/89
>> 1411 root     3333M 3253M cpu29    0    0   7:31:31 1.9% java/87
>> 3461 webservd   40M   32M sleep    0    0   0:00:03 0.3% httpd/1
>> 3460 webservd   36M   26M sleep    0    0   0:00:03 0.3% httpd/1
>> 3462 webservd   36M   26M sleep    0    0   0:00:03 0.3% httpd/1
>> 3457 webservd   32M   27M cpu24    0    0   0:00:02 0.3% httpd/1
>> 1423 root     3333M 3256M sleep    0    0   7:00:01 0.2% java/88
>> 3348 webservd   40M   32M sleep    0    0   0:00:04 0.2% httpd/1
>>  995 root     3536K 3072K sleep  100    -   0:00:46 0.1% cpustat/33
>> 1360 webservd   43M   35M sleep    0    0   0:00:14 0.1% httpd/1
>> 1337 webservd   43M   35M sleep    0    0   0:00:13 0.1% httpd/1
>> 3559 webservd   13M   11M cpu20    0    0   0:00:00 0.1% hgwebdir.cgi/1
>>  883 root     3848K 3832K cpu25    0    0   0:00:13 0.1% prstat/1
>> 1011 webservd   43M   36M sleep    0    0   0:00:15 0.1% httpd/1
>>   77 webservd 9016K 7832K sleep    0    0   0:16:18 0.1% memcached/1
>> Total: 166 processes, 1525 lwps, load averages: 10.00, 10.20, 10.03
>>
>> TIA,
>> fdo
>>
>> ---------------------------------------------------------------------
>> The official User-To-User support forum of the Apache HTTP Server 
>> Project.
>> See <URL:http://httpd.apache.org/userslist.html> for more info.
>> To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
>>   "   from the digest: users-digest-unsubscribe@httpd.apache.org
>> For additional commands, e-mail: users-help@httpd.apache.org
>>
>
>
> ---------------------------------------------------------------------
> The official User-To-User support forum of the Apache HTTP Server 
> Project.
> See <URL:http://httpd.apache.org/userslist.html> for more info.
> To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
>   "   from the digest: users-digest-unsubscribe@httpd.apache.org
> For additional commands, e-mail: users-help@httpd.apache.org
>


---------------------------------------------------------------------
The official User-To-User support forum of the Apache HTTP Server Project.
See <URL:http://httpd.apache.org/userslist.html> for more info.
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
   "   from the digest: users-digest-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org


Mime
View raw message