Thanks a lot. I will look into your suggestion and retry with correct option


On Fri, Sep 30, 2016 at 10:46 AM, Willy Tarreau <[email protected]> wrote:

> Hi,
>
> On Fri, Sep 30, 2016 at 09:09:07AM +0530, Rajesh Mahajan wrote:
> > Hi
> > We are using haproxy as loadbalancer also want to use it as ssl terminal
> > but we are not getting performance as we get though nginx system. Please
> > suggest the best configuration to achive the same.I have tried all ssl
> > optimization on both haproxy and nginx.
> > Below are performance stats
>
> Your test reports many errors for both products and incorrect test
> conditions, so in practice until you can either :
>   - demonstrate that these errors are expected
>   - solve them
>
> you can't tell anything about "performance" (which seems low for both
> products, as both nginx and haproxy are known for being much faster
> than this).
>
> Look :
>
> > =========== NGINX PERFORMACE ====================
> >
> > httperf --hog --timeout=5 --client=0/1 --server=192.168.57.30
> --port=8443
> > --uri=/index.html --http-version=1.0 --rate=500 --send-buffer=4096
> > --recv-buffer=16384 --ssl --ssl-no-reuse --num-conns=5000 --num-calls=2
> > httperf: warning: open file limit > FD_SETSIZE; limiting max. # of open
> > files to FD_SETSIZE Maximum connect burst length: 19
>
> This warning basically says that it will not be able to handle all
> connections in parallel and that you don't know what will happen.
>
> > Total: connections 4666 requests 9332 replies 9332 test-duration 11.675 s
> >
> > Connection rate: 399.6 conn/s (2.5 ms/conn, <=1022 concurrent
> connections)
>
> This one is quite low and makes me suspect you're running inside a VM,
> in which case your tests probably don't mean anything if other VMs are
> running in parallel on the same machine.
>
> > Errors: total 334 client-timo 0 socket-timo 0 connrefused 0 connreset 0
> > Errors: fd-unavail 334 addrunavail 0 ftab-full 0 other 0
>
> This one means that httperf produced its own error and had to abort some
> connections due to its own configuration, which totally invalidates the
> test.
>
> > ========= HA PROXY PERFORMACE ==============
> >
> > httperf --hog --timeout=5 --client=0/1 --server=192.168.57.30 --port=443
> > --uri=/index.html --http-version=1.0 --rate=500 --send-buffer=4096
> > --recv-buffer=16384 --ssl --ssl-no-reuse --num-conns=5000 --num-calls=2
> > httperf: warning: open file limit > FD_SETSIZE; limiting max. # of open
> > files to FD_SETSIZE Maximum connect burst length: 394
>
> Here same warning again with a larger max conn burst length (why ?)
>
> > Total: connections 1666 requests 2933 replies 1267 test-duration 16.194 s
> >
> > Connection rate: 102.9 conn/s (9.7 ms/conn, <=1022 concurrent
> connections)
>
> This performance level becomes ridiculous.
>
> > Errors: total 5000 client-timo 399 socket-timo 0 connrefused 0 connreset
> > 1267 Errors: fd-unavail 3334 addrunavail 0 ftab-full 0 other 0
>
> And here you can see that 2/3 of the connections failed due to
> configuration
> issues on httperf itself.
>
> Also a few comments below :
>
> >     ssl-default-bind-ciphers
> > ECDH+AESGCM:DH+AESGCM:ECDH+AES256:DH+AES256:ECDH+AES128:
> DH+AES:ECDH+3DES:DH+3DES:RSA+AESGCM:RSA+AES:RSA+3DES:!aNULL:!MD5:!DSS
>
> Also I seem to remember that ECDH is different from ECDHE and was slowing
> down the handshake. You should definitely check Mozilla's generator below
> to ensure that your nginx and haproxy configs are optimal :
>
>    https://mozilla.github.io/server-side-tls/ssl-config-generator/
>
> > ============== MY SYSTEM==========
> >
> > 8 core 8 GM RAM
> >
> >               sign    verify    sign/s verify/s
> >
> > rsa 512 bits  0.000157s 0.000013s 6380.2 74831.5
> > rsa 1024 bits 0.000782s 0.000040s 1278.0 24937.4
> > rsa 2048 bits 0.004854s 0.000142s 206.0 7065.4
> > rsa 4096 bits 0.034281s 0.000537s 29.2 1863.5
>
> Here it's quite low... Very low... Make me think of an embedded
> device, but the 8 GB RAM invalidates this option, hence I suspect
> it's a limited VM. Here's what I have here on my home PC, it's
> 41 times faster for half the core count, so surely there's
> something wrong on your setup (!) :
>
> - single core :
>
>                   sign    verify    sign/s verify/s
> rsa  512 bits 0.000034s 0.000002s  29675.7 519937.1
> rsa 1024 bits 0.000072s 0.000004s  13851.6 222222.3
> rsa 2048 bits 0.000466s 0.000014s   2143.9  73890.8
> rsa 4096 bits 0.003133s 0.000048s    319.2  21014.7
>
> - 4 cores :
>
> rsa  512 bits 0.000008s 0.000000s 117647.1 2000000.0
> rsa 1024 bits 0.000018s 0.000001s  55555.6 950000.0
> rsa 2048 bits 0.000118s 0.000003s   8491.2 285714.3
> rsa 4096 bits 0.000793s 0.000012s   1261.0  83333.3
>
> Hoping this helps,
> Willy
>

Reply via email to