Thanks for your reply!--On 9. November 2009 07:24:22 -0500 Adam Tauno Williams <awilliam@xxxxxxxxxxxxx> wrote:
We run a ~500-600 connection Cyrus in ESX. But that isn't ~2,500. There are so many variables here.
True ...
How much I/O your hypervisor can absorb depends on the same thing as a metal system - the underlying I/O subsystem.
We have a 2 Gbit IBM SVC-SAN connected with Fibrechannel over Cisco MDS 9509 switches. So far that's been sufficient, but we will probably upgrade that as well in the forseeable future. We will use the same SAN devices as raw devices under VMware. I've successfully done that on test installations.
So far with ESX I've seen that the I/O subsystem bottlenecks before there is any 'hard limit' in the hypervisor; just like a metal system bottlenecks before any 'hard limit' in a [modern/recent] OS. The real problem with virtualization is that it becomes almost trivial to over-subscribe the system,
Exactly.
it works great so long as only a few systems load spike at once - when seven or eight servers load spike at the same time things get ugly and the poor sys-admin's cell phone gets inundated with SMS alerts.
So the real task would be to define appropriate reservations. But it's good to know that someone actually uses such a system in production, even though it's lower-sized than ours.
--
.:.Sebastian Hagedorn - RZKR-R1 (Gebäude 52), Zimmer 18.:.
.:.Regionales Rechenzentrum (RRZK).:.
.:.Universität zu Köln / Cologne University - ✆ +49-221-478-5587.:.Attachment:
p7s6H4ah7aCiI.p7s
Description: S/MIME cryptographic signature
---- Cyrus Home Page: http://cyrusimap.web.cmu.edu/ Cyrus Wiki/FAQ: http://cyrusimap.web.cmu.edu/twiki List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
