Re: [users@httpd] Poor Load Balancer performance

[Jim Albert]

Some HTTP load balancers perform additional duties such as acting as a 
web application firewall (WAF). The load balancer/WAF helps to weed out 
attacks (and balance load) leaving the back end servers focusing on what 
are seen as legitimate requests... each portion of the reverse proxy 
environment focusing on what it does best. Load balancing WAFs are a 
common part of the HTTP environment where security is a requirement.


Jim

On 5/27/2021 8:02 AM, Daniel Ferradal wrote:
 if you have thousands of connections you need to have more servers 
and then a balancer will help spreading the load (not increasing 
performance) among all the servers, or add bigger figures to your mpm 
settings in your single server (if the hardware will be able to cope 
with it).


If you have thousand of connections and just 800 workers, then it is 
logical that it clogs.


Supposing 800 connections would be enough for peaks if you want no 
clogging with prefork because load is quite variable it is better to 
specify a higher minspareservers value (more workers readily 
available, not have to be spawned, less cpu usage because less 
processes have to be spawned).


I would try to go for event asap though, much better to have to spawn 
8 processes with 100 threads than 800 processes.


El jue, 27 may 2021 a las 13:35, Marc Serra (>) escribió:


Thank's Eric, I understand.

But if configuring a load balancer I'm adding another hop with
reducing the performance, seems that is not the best way to
improve this performance, correct?

As I explained before, it's only a test setup before move to a
real environment.

Now I have a single server (8 cores, 32GB RAM with prefork) and
the following apache tunning. In some moments, I have thousands of
concurrent connections that collapse the server and I thought the
best way to fix this was by setting up a load balancer. Isn't it?

ServerLimit 800
StartServers 10
MinSpareServers 200
MaxSpareServers 400
MaxRequestsWorkers 800
MaxConnectionsPerChild 1

Missatge de Eric Covener mailto:cove...@gmail.com>> del dia dj., 27 de maig 2021 a les 13:15:

On Thu, May 27, 2021 at 7:06 AM Marc Serra mailto:mse...@manxa.com>> wrote:
> I has inverted the results! I'm sorry!

I see -- I think this is a case where you aren't taxing either
backend
and you've just added another hop.
This is probably especially true when testing static files.

-
To unsubscribe, e-mail: users-unsubscr...@httpd.apache.org

For additional commands, e-mail: users-h...@httpd.apache.org




-- 
Marc Serra

Organització i Sistemes


/Manxa 1876, S.L. /
/Ctra. Les Tries, 85. 17800 Olot (Girona)///
//Tel. 972 27 45 30 Fax 972 27 45 32//

* Manxa Industrial | *Coneix més aquí 

* Manxa Ferros | *Coneix més aquí 

* Manxa Ferreteria i Parament de la Llar | *Coneix més aquí


**

El contingut d’aquest correu electrònic i els seus annexos és
estrictament confidencial. En el cas que no siguis el destinatari
i hagis rebut aquest missatge per error, preguem que ho comuniquis
al remitent i procedeixis a la seva eliminació, sense difondre,
emmagatzemar o copiar el seu contingut.Imprimeix aquest correu
només si és necessari.

El contenido de este correo electrónico y sus anexos es
estrictamente confidencial. En el caso de que no seas el
destinatario y hayas recibido este mensaje por error, rogamos lo
comuniques al remitente y procedas a su eliminación, sin difundir,
almacenar o copiar su contenido.Imprimir este correo solo si es
necesario.

The content of this email and its attachments is strictly
confidential. If you are not the recipient and you have received
this message by mistake, please notify the sender and proceed to
its elimination, without spreading, storing or copying its
content. Print this email only if necessary.

Le contenu de cet e-mail et de ses pièces jointes est strictement
confidentiel. Dans le cas où vous n'êtes pas le destinataire et
avez reçu ce message par erreur, veuillez en informer l'expéditeur
et procéder à sa suppression, sans diffuser, stocker ou copier son
contenu.Imprimez cet e-mail uniquement si nécessaire.



--
Daniel Ferradal
HTTPD Project
#httpd help at Libera.Chat

Re: [users@httpd] Poor Load Balancer performance

[Marc Serra]

Thank's for your reply Eric.

 if you have thousands of connections you need to have more servers and
> then a balancer will help spreading the load (not increasing performance)
> among all the servers, or add bigger figures to your mpm settings in your
> single server (if the hardware will be able to cope with it).
>
> If you have thousand of connections and just 800 workers, then it is
> logical that it clogs.
>
> Supposing 800 connections would be enough for peaks if you want no
> clogging with prefork because load is quite variable it is better to
> specify a higher minspareservers value (more workers readily available, not
> have to be spawned, less cpu usage because less processes have to be
> spawned).
>
> I would try to go for event asap though, much better to have to spawn 8
> processes with 100 threads than 800 processes.
>

Daniel, after your words, I think it's important to move to event too. I
will take a look. Thank's.

-- 

Manxa 
1876, S.L. *
Ctra. 
Les Tries, 85. 17800 Olot (Girona)**Tel. 972 27 
45 30 Fax 972 27 45 32*


* Manxa Industrial | *Coneix
més aquí 




* Manxa Ferros | *Coneix
més aquí 




* Manxa Ferreteria i Parament de la Llar | 
*Coneix
més aquí 


**

-- 


El contingut d’aquest correu electrònic i els seus annexos és 
estrictament confidencial. En el cas que no siguis el destinatari i hagis 
rebut aquest missatge per error, preguem que ho comuniquis al remitent i 
procedeixis a la seva eliminació, sense difondre, emmagatzemar o copiar el 
seu contingut. Imprimeix aquest correu només si és necessari.

El contenido 
de este correo electrónico y sus anexos es estrictamente confidencial. En 
el caso de que no seas el destinatario y hayas recibido este mensaje por 
error, rogamos lo comuniques al remitente y procedas a su eliminación, sin 
difundir, almacenar o copiar su contenido. Imprimir este correo solo si es 
necesario.

The content of this email and its attachments is strictly 
confidential. If you are not the recipient and you have received this 
message by mistake, please notify the sender and proceed to its 
elimination, without spreading, storing or copying its content. Print this 
email only if necessary.

Le contenu de cet e-mail et de ses pièces jointes 
est strictement confidentiel. Dans le cas où vous n'êtes pas le 
destinataire et avez reçu ce message par erreur, veuillez en informer 
l'expéditeur et procéder à sa suppression, sans diffuser, stocker ou copier 
son contenu. Imprimez cet e-mail uniquement si nécessaire.

Re: [users@httpd] Poor Load Balancer performance

[Daniel Ferradal]

 if you have thousands of connections you need to have more servers and
then a balancer will help spreading the load (not increasing performance)
among all the servers, or add bigger figures to your mpm settings in your
single server (if the hardware will be able to cope with it).

If you have thousand of connections and just 800 workers, then it is
logical that it clogs.

Supposing 800 connections would be enough for peaks if you want no clogging
with prefork because load is quite variable it is better to specify a
higher minspareservers value (more workers readily available, not have to
be spawned, less cpu usage because less processes have to be spawned).

I would try to go for event asap though, much better to have to spawn 8
processes with 100 threads than 800 processes.

El jue, 27 may 2021 a las 13:35, Marc Serra () escribió:

> Thank's Eric, I understand.
>
> But if configuring a load balancer I'm adding another hop with reducing
> the performance, seems that is not the best way to improve this
> performance, correct?
>
> As I explained before, it's only a test setup before move to a real
> environment.
>
> Now I have a single server (8 cores, 32GB RAM with prefork) and the
> following apache tunning. In some moments, I have thousands of concurrent
> connections that collapse the server and I thought the best way to fix this
> was by setting up a load balancer. Isn't it?
>
> ServerLimit 800
> StartServers 10
> MinSpareServers 200
> MaxSpareServers 400
> MaxRequestsWorkers 800
> MaxConnectionsPerChild 1
>
> Missatge de Eric Covener  del dia dj., 27 de maig 2021
> a les 13:15:
>
>> On Thu, May 27, 2021 at 7:06 AM Marc Serra  wrote:
>> > I has inverted the results! I'm sorry!
>>
>> I see -- I think this is a case where you aren't taxing either backend
>> and you've just added another hop.
>> This is probably especially true when testing static files.
>>
>> -
>> To unsubscribe, e-mail: users-unsubscr...@httpd.apache.org
>> For additional commands, e-mail: users-h...@httpd.apache.org
>>
>>
>
> --
> Marc Serra
> Organització i Sistemes
>
>
>
> *Manxa 1876, S.L. Ctra. Les Tries, 85. 17800 Olot (Girona)*
> *Tel. 972 27 45 30 Fax 972 27 45 32*
>
> * Manxa Industrial | *Coneix més aquí 
>
> * Manxa Ferros | *Coneix més aquí 
>
> * Manxa Ferreteria i Parament de la Llar | *Coneix més aquí
> 
>
>
> El contingut d’aquest correu electrònic i els seus annexos és estrictament
> confidencial. En el cas que no siguis el destinatari i hagis rebut aquest
> missatge per error, preguem que ho comuniquis al remitent i procedeixis a
> la seva eliminació, sense difondre, emmagatzemar o copiar el seu contingut.
> Imprimeix aquest correu només si és necessari.
>
> El contenido de este correo electrónico y sus anexos es estrictamente
> confidencial. En el caso de que no seas el destinatario y hayas recibido
> este mensaje por error, rogamos lo comuniques al remitente y procedas a su
> eliminación, sin difundir, almacenar o copiar su contenido. Imprimir este
> correo solo si es necesario.
>
> The content of this email and its attachments is strictly confidential. If
> you are not the recipient and you have received this message by mistake,
> please notify the sender and proceed to its elimination, without spreading,
> storing or copying its content. Print this email only if necessary.
>
> Le contenu de cet e-mail et de ses pièces jointes est strictement
> confidentiel. Dans le cas où vous n'êtes pas le destinataire et avez reçu
> ce message par erreur, veuillez en informer l'expéditeur et procéder à sa
> suppression, sans diffuser, stocker ou copier son contenu. Imprimez cet
> e-mail uniquement si nécessaire.
>


-- 
Daniel Ferradal
HTTPD Project
#httpd help at Libera.Chat

Re: [users@httpd] Poor Load Balancer performance

[Eric Covener]

On Thu, May 27, 2021 at 7:41 AM Marc Serra  wrote:

> Thank's Eric, I understand.
>
> But if configuring a load balancer I'm adding another hop with reducing
> the performance, seems that is not the best way to improve this
> performance, correct?
>
> As I explained before, it's only a test setup before move to a real
> environment.
>
> Now I have a single server (8 cores, 32GB RAM with prefork) and the
> following apache tunning. In some moments, I have thousands of concurrent
> connections that collapse the server and I thought the best way to fix this
> was by setting up a load balancer. Isn't it?
>
> ServerLimit 800
> StartServers 10
> MinSpareServers 200
> MaxSpareServers 400
> MaxRequestsWorkers 800
> MaxConnectionsPerChild 1
>

"load balancer" is an overloaded term.
A HTTP reverse proxy is one way to scale to multiple servers. There are
ways that are lighter weight at runtime like a VIP or NAT forwarding.

But it may be moot for your real workload.

Re: [users@httpd] Poor Load Balancer performance

[Marc Serra]

Thank's Eric, I understand.

But if configuring a load balancer I'm adding another hop with reducing the
performance, seems that is not the best way to improve this performance,
correct?

As I explained before, it's only a test setup before move to a real
environment.

Now I have a single server (8 cores, 32GB RAM with prefork) and the
following apache tunning. In some moments, I have thousands of concurrent
connections that collapse the server and I thought the best way to fix this
was by setting up a load balancer. Isn't it?

ServerLimit 800
StartServers 10
MinSpareServers 200
MaxSpareServers 400
MaxRequestsWorkers 800
MaxConnectionsPerChild 1

Missatge de Eric Covener  del dia dj., 27 de maig 2021 a
les 13:15:

> On Thu, May 27, 2021 at 7:06 AM Marc Serra  wrote:
> > I has inverted the results! I'm sorry!
>
> I see -- I think this is a case where you aren't taxing either backend
> and you've just added another hop.
> This is probably especially true when testing static files.
>
> -
> To unsubscribe, e-mail: users-unsubscr...@httpd.apache.org
> For additional commands, e-mail: users-h...@httpd.apache.org
>
>

-- 
Marc Serra
Organització i Sistemes

-- 

Manxa 
1876, S.L. *
Ctra. 
Les Tries, 85. 17800 Olot (Girona)**Tel. 972 27 
45 30 Fax 972 27 45 32*


* Manxa Industrial | *Coneix
més aquí 




* Manxa Ferros | *Coneix
més aquí 




* Manxa Ferreteria i Parament de la Llar | 
*Coneix
més aquí 


**

-- 


El contingut d’aquest correu electrònic i els seus annexos és 
estrictament confidencial. En el cas que no siguis el destinatari i hagis 
rebut aquest missatge per error, preguem que ho comuniquis al remitent i 
procedeixis a la seva eliminació, sense difondre, emmagatzemar o copiar el 
seu contingut. Imprimeix aquest correu només si és necessari.

El contenido 
de este correo electrónico y sus anexos es estrictamente confidencial. En 
el caso de que no seas el destinatario y hayas recibido este mensaje por 
error, rogamos lo comuniques al remitente y procedas a su eliminación, sin 
difundir, almacenar o copiar su contenido. Imprimir este correo solo si es 
necesario.

The content of this email and its attachments is strictly 
confidential. If you are not the recipient and you have received this 
message by mistake, please notify the sender and proceed to its 
elimination, without spreading, storing or copying its content. Print this 
email only if necessary.

Le contenu de cet e-mail et de ses pièces jointes 
est strictement confidentiel. Dans le cas où vous n'êtes pas le 
destinataire et avez reçu ce message par erreur, veuillez en informer 
l'expéditeur et procéder à sa suppression, sans diffuser, stocker ou copier 
son contenu. Imprimez cet e-mail uniquement si nécessaire.

Re: [users@httpd] Poor Load Balancer performance

[Eric Covener]

On Thu, May 27, 2021 at 7:06 AM Marc Serra  wrote:
> I has inverted the results! I'm sorry!

I see -- I think this is a case where you aren't taxing either backend
and you've just added another hop.
This is probably especially true when testing static files.

-
To unsubscribe, e-mail: users-unsubscr...@httpd.apache.org
For additional commands, e-mail: users-h...@httpd.apache.org

Re: [users@httpd] Poor Load Balancer performance

[Marc Serra]

Oh shit! copy/paste error!!!

I has inverted the results! I'm sorry!

the real ones...

Results when I run 'ab  -k -n 1 -c 1000 http://vm_host_ip_address:8010/'
with NO balancer configured...
 Finished 1 requests
 Server Software:Apache/2.4.41
 Server Hostname:192.168.68.210
 Server Port:8010
 Document Path:  /
 Document Length:97098 bytes
 Concurrency Level:  1000
 Time taken for tests:   15.527 seconds
 Complete requests:  1
 Failed requests:834
   (Connect: 0, Receive: 0, Length: 834, Exceptions: 0)
 Keep-Alive requests:9166
 Total transferred:  892851836 bytes
 HTML transferred:   89268 bytes
 Requests per second:644.04 [#/sec] (mean)
 Time per request:   1552.701 [ms] (mean)
 Time per request:   1.553 [ms] (mean, across all concurrent requests)
 Transfer rate:  56155.42 [Kbytes/sec] received
 Connection Times (ms)
  min  mean[+/-sd] median   max
 Connect:0   22 146.6  01038
 Processing: 0  657 1668.1122   13453
 Waiting:0  752 1883.8121   13453
 Total:  0  679 1763.1122   14476
 Percentage of the requests served within a certain time (ms)
  50%122
  66%234
  75%330
  80%402
  90%   1657
  95%   4987
  98%   7221
  99%   8733
  100%  14476 (longest request)
 Finished 1 requests



Results when I run 'ab  -k -n 1 -c 1000 http://vm_host_ip_address:8010/'
with (YES) balancer...
 Finished 1 requests
 Server Software:Apache/2.4.41
 Server Hostname:192.168.68.210
 Server Port:8010
 Document Path:  /
 Document Length:97098 bytes
 Concurrency Level:  1000
 Time taken for tests:   24.029 seconds
 Complete requests:  1
 Failed requests:5526
(Connect: 0, Receive: 0, Length: 5526, Exceptions: 0)
 Keep-Alive requests:8938
 Total transferred:  870844342 bytes
 HTML transferred:   868062523 bytes
 Requests per second:416.16 [#/sec] (mean)
 Time per request:   2402.908 [ms] (mean)
 Time per request:   2.403 [ms] (mean, across all concurrent requests)
 Transfer rate:  35391.87 [Kbytes/sec] received
 Connection Times (ms)
  min  mean[+/-sd] median   max
 Connect:0   22 144.2  01039
 Processing: 0 1087 2291.1451   18504
 Waiting:0 1207 2615.1426   18470
 Total:  0 1109 2372.3451   19514
 Percentage of the requests served within a certain time (ms)
  50%451
  66%585
  75%678
  80%764
  90%   1868
  95%   7315
  98%  10098
  99%  11156
  100%  19514 (longest request)
 Finished 1 requests

Missatge de Eric Covener  del dia dj., 27 de maig 2021 a
les 12:51:

> On Thu, May 27, 2021 at 6:22 AM Marc Serra  wrote:
>
>> I got better performance without the load balancer.
>>
>
>
>> Results when I run 'ab  -k -n 1 -c 1000
>> http://vm_host_ip_address:8010/' with no balancer...
>>  Requests per second:416.16 [#/sec] (mean)
>>  Transfer rate:  35391.87 [Kbytes/sec] received
>>
>
>
>
>> Results when I run 'ab  -k -n 1 -c 1000
>> http://vm_host_ip_address:8010/' with balancer configured...
>>  Requests per second:644.04 [#/sec] (mean)
>>  Transfer rate:  56155.42 [Kbytes/sec] received
>>
>
>
> It doesn't look like the no-proxy case did better.  Which metric were you
> looking at?
>
> + a few cautions:
> - If you aren't taxing the system beyond what a single backend could
> handle, it can easily be slower by adding a proxy.
> - A layer 7 proxy can easily be more overhead than a basic VIP / LB / IP
> sprayer if there's nothing actually being offloaded in the proxy.
>
>


-- 
Marc Serra
Organització i Sistemes

-- 

Manxa 
1876, S.L. *
Ctra. 
Les Tries, 85. 17800 Olot (Girona)**Tel. 972 27 
45 30 Fax 972 27 45 32*


* Manxa Industrial | *Coneix
més aquí 




* Manxa Ferros | *Coneix
més aquí 




* Manxa Ferreteria i Parament de la Llar | 
*Coneix
més aquí 


**

-- 


El contingut d’aquest correu electrònic i els seus annexos és 
estrictament confidencial. En el cas que no siguis el destinatari i hagis 
rebut aquest missatge per error, preguem que ho comuniquis al remitent i 
procedeixis a la seva eliminació, sense difondre, emmagatzemar o copiar el 
seu contingut. Imprimeix aquest correu només si és necessari.

El contenido 
de este correo electrónico y sus anexos es estrictamente confidencial. En 
el caso de que no seas el destinatario y hayas recibido este mensaje por 
error, rogamos lo comuniques al remitente y procedas a su eliminación, sin 
difundir, almacenar o copiar su contenido. Imprimir este correo solo si es 
necesario.

The content of this email and its attachments is strictly 
confidential. If you are not the recipient and you have received this 
message by mistake, please notify the sender and proceed to its 
elimination, without 

Re: [users@httpd] Poor Load Balancer performance

[Eric Covener]

On Thu, May 27, 2021 at 6:22 AM Marc Serra  wrote:

> I got better performance without the load balancer.
>


> Results when I run 'ab  -k -n 1 -c 1000
> http://vm_host_ip_address:8010/' with no balancer...
>  Requests per second:416.16 [#/sec] (mean)
>  Transfer rate:  35391.87 [Kbytes/sec] received
>



> Results when I run 'ab  -k -n 1 -c 1000
> http://vm_host_ip_address:8010/' with balancer configured...
>  Requests per second:644.04 [#/sec] (mean)
>  Transfer rate:  56155.42 [Kbytes/sec] received
>


It doesn't look like the no-proxy case did better.  Which metric were you
looking at?

+ a few cautions:
- If you aren't taxing the system beyond what a single backend could
handle, it can easily be slower by adding a proxy.
- A layer 7 proxy can easily be more overhead than a basic VIP / LB / IP
sprayer if there's nothing actually being offloaded in the proxy.

Re: [users@httpd] Poor Load Balancer performance

[Nick Folino]

I would expect that adding anything (including a load balancer) in the path
would increase the time.

On Thu, May 27, 2021 at 6:22 AM Marc Serra  wrote:

> Hi, I'm trying to understand how to load balancing works and testing it's
> benefits, and I'm in a trouble: in my tests (using ab command) I got better
> performance without the load balancer.
>
> I try to describe my environment...
>
> I created three VM on Virtualbox with the same hardware and software
> configuration and redirected ports using NAT:
>  CPU: 1
>  RAM: 1024MB
>  Network: NAT
>  Apache version: 2.4.41
>  Kernel: Linux node1 5.4.0-73-generic #82-Ubuntu SMP Wed Apr 14 17:39:42
> UTC 2021 x86_64 x86_64 x86_64 GNU/Linux
>  OS: Ubuntu server 20.04.2
>
> The machines:
>  central: only acts as a load balancer (not a balancer member), otherwise
> used as 'normal' apache server for results comparison. NAT from 8010 to 80
>  node1: when load balancer is configured in the central machine, acts as
> balancer member. NAT from 8011 to 80
>  node2: when load balancer is configured in the central machine, acts as
> balancer member. NAT from 8012 to 80
>
> Here a list of apache loaded module.
>
> spoiler!!! node1 and node2 have the same modules. central have the same
> modules as nod1 and node2 plus lbmethod_byrequests_module (shared)...
>
> Apache loaded modules in central:
> Loaded Modules:
> core_module (static)
> so_module (static)
> watchdog_module (static)
> http_module (static)
> log_config_module (static)
> logio_module (static)
> version_module (static)
> unixd_module (static)
> access_compat_module (shared)
> alias_module (shared)
> auth_basic_module (shared)
> authn_core_module (shared)
> authn_file_module (shared)
> authz_core_module (shared)
> authz_host_module (shared)
> authz_user_module (shared)
> autoindex_module (shared)
> deflate_module (shared)
> dir_module (shared)
> env_module (shared)
> filter_module (shared)
> lbmethod_byrequests_module (shared)
> mime_module (shared)
> mpm_event_module (shared)
> negotiation_module (shared)
> proxy_module (shared)
> proxy_balancer_module (shared)
> proxy_http_module (shared)
> reqtimeout_module (shared)
> setenvif_module (shared)
> slotmem_shm_module (shared)
> status_module (shared)
>
>
> Apache loaded modules in node1:
> core_module (static)
> so_module (static)
> watchdog_module (static)
> http_module (static)
> log_config_module (static)
> logio_module (static)
> version_module (static)
> unixd_module (static)
> access_compat_module (shared)
> alias_module (shared)
> auth_basic_module (shared)
> authn_core_module (shared)
> authn_file_module (shared)
> authz_core_module (shared)
> authz_host_module (shared)
> authz_user_module (shared)
> autoindex_module (shared)
> deflate_module (shared)
> dir_module (shared)
> env_module (shared)
> filter_module (shared)
> mime_module (shared)
> mpm_event_module (shared)
> negotiation_module (shared)
> proxy_module (shared)
> proxy_balancer_module (shared)
> proxy_http_module (shared)
> reqtimeout_module (shared)
> setenvif_module (shared)
> slotmem_shm_module (shared)
> status_module (shared)
>
> Apache loaded modules in node2:
> core_module (static)
> so_module (static)
> watchdog_module (static)
> http_module (static)
> log_config_module (static)
> logio_module (static)
> version_module (static)
> unixd_module (static)
> access_compat_module (shared)
> alias_module (shared)
> auth_basic_module (shared)
> authn_core_module (shared)
> authn_file_module (shared)
> authz_core_module (shared)
> authz_host_module (shared)
> authz_user_module (shared)
> autoindex_module (shared)
> deflate_module (shared)
> dir_module (shared)
> env_module (shared)
> filter_module (shared)
> mime_module (shared)
> mpm_event_module (shared)
> negotiation_module (shared)
> proxy_module (shared)
> proxy_balancer_module (shared)
> proxy_http_module (shared)
> reqtimeout_module (shared)
> setenvif_module (shared)
> slotmem_shm_module (shared)
> status_module (shared)
>
> Results when I run 'ab  -k -n 1 -c 1000
> http://vm_host_ip_address:8010/' with no balancer...
>  Finished 1 requests
>  Server Software:Apache/2.4.41
>  Server Hostname:192.168.68.210
>  Server Port:8010
>  Document Path:  /
>  Document Length:97098 bytes
>  Concurrency Level:  1000
>  Time taken for tests:   24.029 seconds
>  Complete requests:  1
>  Failed requests:5526
> (Connect: 0, Receive: 0, Length: 5526, Exceptions: 0)
>  Keep-Alive requests:8938
>  Total transferred:  870844342 bytes
>  HTML transferred:   868062523 bytes
>  Requests per second:416.16 [#/sec] (mean)
>  Time per request:   2402.908 [ms] (mean)
>  Time per request:   2.403 [ms] (mean, across all concurrent requests)
>  Transfer rate:  35391.87 [Kbytes/sec] received
>  Connection Times (ms)
>   min  mean[+/-sd] median   max
>  Connect:0   22 144.2  01039
>  Processing: 0 1087 2291.1451   18504
>  

[users@httpd] Poor Load Balancer performance

[Marc Serra]

Hi, I'm trying to understand how to load balancing works and testing it's
benefits, and I'm in a trouble: in my tests (using ab command) I got better
performance without the load balancer.

I try to describe my environment...

I created three VM on Virtualbox with the same hardware and software
configuration and redirected ports using NAT:
 CPU: 1
 RAM: 1024MB
 Network: NAT
 Apache version: 2.4.41
 Kernel: Linux node1 5.4.0-73-generic #82-Ubuntu SMP Wed Apr 14 17:39:42
UTC 2021 x86_64 x86_64 x86_64 GNU/Linux
 OS: Ubuntu server 20.04.2

The machines:
 central: only acts as a load balancer (not a balancer member), otherwise
used as 'normal' apache server for results comparison. NAT from 8010 to 80
 node1: when load balancer is configured in the central machine, acts as
balancer member. NAT from 8011 to 80
 node2: when load balancer is configured in the central machine, acts as
balancer member. NAT from 8012 to 80

Here a list of apache loaded module.

spoiler!!! node1 and node2 have the same modules. central have the same
modules as nod1 and node2 plus lbmethod_byrequests_module (shared)...

Apache loaded modules in central:
Loaded Modules:
core_module (static)
so_module (static)
watchdog_module (static)
http_module (static)
log_config_module (static)
logio_module (static)
version_module (static)
unixd_module (static)
access_compat_module (shared)
alias_module (shared)
auth_basic_module (shared)
authn_core_module (shared)
authn_file_module (shared)
authz_core_module (shared)
authz_host_module (shared)
authz_user_module (shared)
autoindex_module (shared)
deflate_module (shared)
dir_module (shared)
env_module (shared)
filter_module (shared)
lbmethod_byrequests_module (shared)
mime_module (shared)
mpm_event_module (shared)
negotiation_module (shared)
proxy_module (shared)
proxy_balancer_module (shared)
proxy_http_module (shared)
reqtimeout_module (shared)
setenvif_module (shared)
slotmem_shm_module (shared)
status_module (shared)


Apache loaded modules in node1:
core_module (static)
so_module (static)
watchdog_module (static)
http_module (static)
log_config_module (static)
logio_module (static)
version_module (static)
unixd_module (static)
access_compat_module (shared)
alias_module (shared)
auth_basic_module (shared)
authn_core_module (shared)
authn_file_module (shared)
authz_core_module (shared)
authz_host_module (shared)
authz_user_module (shared)
autoindex_module (shared)
deflate_module (shared)
dir_module (shared)
env_module (shared)
filter_module (shared)
mime_module (shared)
mpm_event_module (shared)
negotiation_module (shared)
proxy_module (shared)
proxy_balancer_module (shared)
proxy_http_module (shared)
reqtimeout_module (shared)
setenvif_module (shared)
slotmem_shm_module (shared)
status_module (shared)

Apache loaded modules in node2:
core_module (static)
so_module (static)
watchdog_module (static)
http_module (static)
log_config_module (static)
logio_module (static)
version_module (static)
unixd_module (static)
access_compat_module (shared)
alias_module (shared)
auth_basic_module (shared)
authn_core_module (shared)
authn_file_module (shared)
authz_core_module (shared)
authz_host_module (shared)
authz_user_module (shared)
autoindex_module (shared)
deflate_module (shared)
dir_module (shared)
env_module (shared)
filter_module (shared)
mime_module (shared)
mpm_event_module (shared)
negotiation_module (shared)
proxy_module (shared)
proxy_balancer_module (shared)
proxy_http_module (shared)
reqtimeout_module (shared)
setenvif_module (shared)
slotmem_shm_module (shared)
status_module (shared)

Results when I run 'ab  -k -n 1 -c 1000 http://vm_host_ip_address:8010/'
with no balancer...
 Finished 1 requests
 Server Software:Apache/2.4.41
 Server Hostname:192.168.68.210
 Server Port:8010
 Document Path:  /
 Document Length:97098 bytes
 Concurrency Level:  1000
 Time taken for tests:   24.029 seconds
 Complete requests:  1
 Failed requests:5526
(Connect: 0, Receive: 0, Length: 5526, Exceptions: 0)
 Keep-Alive requests:8938
 Total transferred:  870844342 bytes
 HTML transferred:   868062523 bytes
 Requests per second:416.16 [#/sec] (mean)
 Time per request:   2402.908 [ms] (mean)
 Time per request:   2.403 [ms] (mean, across all concurrent requests)
 Transfer rate:  35391.87 [Kbytes/sec] received
 Connection Times (ms)
  min  mean[+/-sd] median   max
 Connect:0   22 144.2  01039
 Processing: 0 1087 2291.1451   18504
 Waiting:0 1207 2615.1426   18470
 Total:  0 1109 2372.3451   19514
 Percentage of the requests served within a certain time (ms)
  50%451
  66%585
  75%678
  80%764
  90%   1868
  95%   7315
  98%  10098
  99%  11156
  100%  19514 (longest request)
 Finished 1 requests






No changes on vhost configuration file...


ServerAdmin webmaster@localhost
DocumentRoot /var/www/html

ErrorLog