LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: SV: local node and remote nodes - Problem!

To: Andersson Mattias <Mattias.Andersson@xxxxxxx>
Subject: Re: SV: local node and remote nodes - Problem!
Cc: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
From: David Osojnik <david@xxxxxxxxxxx>
Date: Mon, 03 Oct 2005 15:32:18 +0200

Andersson Mattias wrote:

>Hi David.
>
>I think that the checks you do do not properly reflect your real distribution 
>of the trafic. 
>
>  
>
>>-----Ursprungligt meddelande-----
>>Från: David Osojnik [mailto:david@xxxxxxxxxxx]
>>Skickat: den 3 oktober 2005 14:06
>>Till: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
>>Ämne: local node and remote nodes - Problem!
>>
>>Hello,
>>
>>I have a problem with LVS:
>>
>>I have a web server LVS and i have a problem like this:
>>
>>All the REMOTE nodes work GREAT and are distributed nicely!
>>The problem is Local Node! It has FAR Less connections then other nodes!
>>And i would like them to have the same processes each! Local or Remote!
>>
>>if i do:
>>
>>"ps aux |grep httpd |wc -l" on all the nodes i get this:
>>
>>IP : httpd processes
>>10.0.0.1: 29
>>10.0.0.2: 508
>>10.0.0.3: 462
>>10.0.0.4: 500
>>10.0.0.5: 471
>>10.0.0.6: 459
>>
>>    
>>
>
>This just lists current number of running httpd processes.
>  
>
Yes but why do the processes on localhost finish much before the others?

>
>  
>
>>Here is the output of "ipvsadm -L -n":
>>
>>IP Virtual Server version 1.2.0 (size=4096)
>>Prot LocalAddress:Port Scheduler Flags
>>  -> RemoteAddress:Port           Forward Weight ActiveConn InActConn
>>TCP  xxx.xxx.xxx.xxx:80 rr
>>  -> 10.0.0.1:80                Local   100    48         3273
>>  -> 10.0.0.2:80                Route   100    34559      29
>>  -> 10.0.0.3:80                Route   100    34518      42
>>  -> 10.0.0.4:80                Route   100    34605      39
>>  -> 10.0.0.5:80                Route   100    34627      37
>>  -> 10.0.0.6:80                Route   100    34582      36
>>
>>    
>>
>
>This just shows number of active and inactive connections on the diffrent 
>servers. Perhapps, IMHO, the local node can process its incoming connections 
>faster the the direct routed nodes?
>
>
>  
>
Maybe but why? All the servers have the same Hardware and same
installations! And If I start LVS on lets say server 4 (it gets to be
local node) and then Server 4 (localnode) gets MUCH MUCH less processes.

>>I remember a time back when we had RedHat 9 it worked great but now it
>>is not working good!
>>
>>I tried changing the LVS to other servers so i got Local on other
>>servers but the problem is the same with the Local node! (far less
>>connections)
>>
>>Any ideas? Thank You!
>>    
>>
>
>I would try with ipvsadm -L --stats and/or ipvsadm -L --rate
>to see the actual distributions of connections.
>
>Hope that this will show you a more even distribution.
>
>Regards 
>/Mattias
>
>
>  
>
# ipvsadm -L -n --stats
IP Virtual Server version 1.2.0 (size=4096)
Prot LocalAddress:Port               Conns   InPkts  OutPkts  InBytes
OutBytes
  -> RemoteAddress:Port
TCP  xxx.xxx.xxx.xxx:80             85525749  554024K        0  
86259M        0
  -> 10.0.0.101:80                14236269  104509K        0  
14978M        0
  -> 10.0.0.102:80                14254728 90860798        0  
14301M        0
  -> 10.0.0.103:80                14261824 89606530        0  
14245M        0
  -> 10.0.0.104:80                14257988 89650485        0  
14244M        0
  -> 10.0.0.105:80                14256043 89644097        0  
14242M        0
  -> 10.0.0.106:80                14258897 89753309        0  
14247M        0

# ipvsadm -L -n --rate
IP Virtual Server version 1.2.0 (size=4096)
Prot LocalAddress:Port                 CPS    InPPS   OutPPS    InBPS  
OutBPS
  -> RemoteAddress:Port
TCP  xxx.xxx.xxx.xxx:80                  206     1346        0  
202095        0
  -> 10.0.0.101:80                      34      243        0   
34077        0
  -> 10.0.0.102:80                      34      232        0   
35144        0
  -> 10.0.0.103:80                      34      210        0   
33754        0
  -> 10.0.0.104:80                      34      222        0   
32918        0
  -> 10.0.0.105:80                      34      214        0   
32797        0
  -> 10.0.0.106:80                      34      224        0   
33404        0

Seems like the LVS is accepting connections OK but then when waiting for
FIN packet something is wrong? Like the connections on Remote hosts dont
end? (they wait for timeout)?

Should it be that for HTTP connections there should be MORE InnActConn?
and LESS ActiveCon?

>>I use:
>>- RedHat Enterprise Server 4
>>- ipvsadm v1.24 2003/06/07 (compiled with popt and IPVS v1.2.0)
>>- Keepalived v1.1.11 (04/13,2005)
>>- 2.6.9-11.ELsmp #1 SMP Fri May 20 18:26:27 EDT 2005 i686 i686 i386
>>GNU/Linux
>>- scheduler is rr or wrr or wlc (same problem on all)
>>    
>>

<Prev in Thread] Current Thread [Next in Thread>