[OpenAFS] Re: recommended hardware for new production environment
Jonathan Nilsson
jnilsson@uci.edu
Fri, 19 Mar 2010 16:15:30 -0700
--0016e6d99d7e33075b04822f87ff
Content-Type: text/plain; charset=UTF-8
> Do you have an estimate on the number of client machines? How many
> database servers?
>
We will probably have about 500 client machines as well.
>
> Though even without that, I'm inclined to say that that is really really
> overly sufficient (overkill) for database servers for a cell of that
> size. Though that highly depends on usage, I'd imagine hitting
> fileserver's bottlenecks way before dbservers'.
>
I anticipate having fairly low file server usage (in terms of bandwidth
requirements, frequency of hits, etc). Most of it will be faculty and staff
websites, documents, backups, etc.
Any reason you're specifically asking about dbserver hardware specs, not
> fileserver specs?
>
With the file servers, I figure that we can more easily expand our capacity
by moving volumes. I was more concerned with getting the dbservers right
the first time because it seems quite a bit more complicated to add/remove
database servers (not something you'd want to do on a whim) and can even
cause service downtime if you have to replace the dbserver with the lowest
IP (from step 6 on this page
http://docs.openafs.org/QuickStartUnix/ch03s02.html#LIWQ117).
If I recall correctly, the ACM@UIUC cell is similar in size (perhaps a
> bit smaller) to what you describe. (A few hundred users, around 50 or 60
> project volumes, storage I think is a TB or two. Nearly all web data for
> them is served directly from AFS.) They are currently running dbservers
> on VMs with 256M of RAM. The VM hosts only have between 512M and 1.5G of
> RAM total, I think, and are PIIIs. I don't recall any problems with
> them, but your client load may be quite a bit higher. Just a data point.
>
Thanks! That is great news to hear this comparison. Our ESX hosts range
from older Dual 2.6GHz Opteron systems with 16GB RAM to new Dual Six-core
3GHz Xeon systems with 32GB RAM. Of course there are lots of other VMs
already running on this infrastructure...
--
Jonathan Nilsson, jnilsson@uci.edu
Social Sciences Computing Services
949.824.1536, SSPA 4110, UC Irvine
--0016e6d99d7e33075b04822f87ff
Content-Type: text/html; charset=UTF-8
Content-Transfer-Encoding: quoted-printable
<br>
<div class=3D"gmail_quote"><blockquote class=3D"gmail_quote" style=3D"margi=
n: 0pt 0pt 0pt 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-le=
ft: 1ex;">Do you have an estimate on the number of client machines? How man=
y<br>
database servers?<br></blockquote><div><br>We will probably have about 500 =
client machines as well.<br>=C2=A0</div><blockquote class=3D"gmail_quote" s=
tyle=3D"margin: 0pt 0pt 0pt 0.8ex; border-left: 1px solid rgb(204, 204, 204=
); padding-left: 1ex;">
<br>
Though even without that, I'm inclined to say that that is really reall=
y<br>
overly sufficient (overkill) for database servers for a cell of that<br>
size. Though that highly depends on usage, I'd imagine hitting<br>
fileserver's bottlenecks way before dbservers'.<br></blockquote><di=
v><br>I anticipate having fairly low file server usage (in terms of bandwid=
th requirements, frequency of hits, etc).=C2=A0 Most of it will be faculty =
and staff websites, documents, backups, etc.<br>
<br></div><blockquote class=3D"gmail_quote" style=3D"margin: 0pt 0pt 0pt 0.=
8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;">Any rea=
son you're
specifically asking about dbserver hardware specs, not fileserver specs?<br=
></blockquote><div><br>With the file servers, I figure that we can more eas=
ily expand our capacity by moving volumes.=C2=A0 I was more concerned with =
getting the dbservers right the first time because it seems quite a bit mor=
e complicated to add/remove database servers (not something you'd want =
to do on a whim) and can even cause service downtime if you have to replace=
the dbserver with the lowest IP (from step 6 on this page <a href=3D"http:=
//docs.openafs.org/QuickStartUnix/ch03s02.html#LIWQ117">http://docs.openafs=
.org/QuickStartUnix/ch03s02.html#LIWQ117</a>).<br>
<br></div><blockquote class=3D"gmail_quote" style=3D"margin: 0pt 0pt 0pt 0.=
8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;">
If I recall correctly, the ACM@UIUC cell is similar in size (perhaps a<br>
bit smaller) to what you describe. (A few hundred users, around 50 or 60<br=
>
project volumes, storage I think is a TB or two. Nearly all web data for<br=
>
them is served directly from AFS.) They are currently running dbservers<br>
on VMs with 256M of RAM. The VM hosts only have between 512M and 1.5G of<br=
>
RAM total, I think, and are PIIIs. I don't recall any problems with<br>
them, but your client load may be quite a bit higher. Just a data point.<br=
></blockquote><div><br>Thanks!=C2=A0 That is great news to hear this compar=
ison.=C2=A0 Our ESX hosts range from older Dual 2.6GHz Opteron systems with=
16GB RAM to new Dual Six-core 3GHz Xeon systems with 32GB RAM.=C2=A0=C2=A0=
Of course there are lots of other VMs already running on this infrastructu=
re...<br>
<br></div></div>--<br>Jonathan Nilsson, <a href=3D"mailto:jnilsson@uci.edu"=
>jnilsson@uci.edu</a><br>Social Sciences Computing=20
Services<br>949.824.1536, SSPA 4110, UC Irvine<br>
<br>
--0016e6d99d7e33075b04822f87ff--