maas-devel team mailing list archive

[John Arbash Meinel <john@xxxxxxxxxxxxxxxxx>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 10/24/2012 02:33 PM, Gavin Panella wrote:
> On 24 October 2012 13:15, John Arbash Meinel
> <john@xxxxxxxxxxxxxxxxx> wrote:
>> With 16 Cluster Controllers, we can achieve CPU saturation on
>> the Database with a shared Appserver (19.2s rebuild). Moving the
>> appserver can relieve some of this overhead, but it adds
>> bandwidth considerations between the DB and Appserver. With
>> 'hostnossl' separating the appserver actually slows down
>> rebuilding tags because the network becomes saturated at 900+Mbps
>> (24.6s). Switching to 'hostssl' increases some CPU overhead on
>> the Database machine, but is a net win (17.7s).
>> 
>> Enabling a second Appserver is not directly a benefit (19.1s). As
>> a quirk of our topology, you can run 1 appserver in 'low cpu
>> high bandwidth nossl' mode, and one in 'higher cpu low bandwidth
>> ssl' mode, and maximize the DB's available resources (15.1s).
> 
> What do the "hostnossl", "hostssl", "low cpu high bandwidth nossl"
> and "higher cpu low bandwidth ssl" phrases refer to? PostgreSQL?
> 

Right. I had a lot of details in the email, and then I stripped them
out "for clarity" :).

So if you supply 'hostssl' in postgres, when a client connects, it
encrypts (and compresses) the stream. Which uses less bandwidth, but
more CPU.

John
=:->

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://www.enigmail.net/

iEYEARECAAYFAlCH8oEACgkQJdeBCYSNAAOAlwCfaqi+CtOqSDnLc8vHuBH7LA3u
lPAAoMtJLWfDAQGq9NHlxM5RBR+1KYQ+
=lD/7
-----END PGP SIGNATURE-----