On Thu, 25 Jun 2009 07:41:42 -0600, Matt Kassawara battery@writeme.com wrote:

I'm running mirror mode on a pair of VMs and haven't noticed any issues with replication. If you're using VMware ESX/ESXi 3.5, I recommend

configuring

Linux guests with the kernel boot parameter "clocksource=acpi_pm" and configuring VMware Tools to handle clock management in the guest. Verify NTP isn't running in the guest. Also, making sure the Linux kernel periodic interrupt timer runs at 100 Hz instead of 250 or 1000 Hz can reduce

timing

issues. VMware provides some documentation on best practices for performance tuning I'd recommend reading.

Yes, I have read these (but was using VM workstation). Also, My experimentation was during last summer 2008 and at the same period VMWare switch from "DO NOT USE NTP in Guest, instead VmwareTools" to "Use NTP" without any explanation ...

The main trouble was when modifying on both side "too quickly".

Other testings with differents applications show up that virtualization is no good for timing-picky application (but it is going off-topic ;)

Thanks for the sharing.

Sincerely yours, Mathieu.

On Thu, Jun 25, 2009 at 12:16 AM, Mathieu MILLET ldap@htam.net wrote:

...

On Tue, 23 Jun 2009 09:36:23 +0100, Fernando Silva fernandolrsilva@gmail.com wrote:

...

Hi all,

Hi

...

I'm trying to setup a Mirror Mode configuration using 2 machines running OpenLDAP v2.4.11 on both. To make things easy both machines are on the same (private) network.

[snip] mirrormode on

I'm using Ubuntu 8.10 inside a VMWare image, but both images are running on independent machines and the network on both is Bridge Mode.

...

From previous experiment, especially with OpenLDAP 2.4.11, I have very

hard time with making OpenLDAP N-way replication working on VM, especially VMWare-based ones.

The main problem was with the management of the system clock, where VMWare was making the clock slower and then make a "jump" to compensate the slowiness. So, I got a lot of line "new modification with older timer value" (can't remember the precise log line, though).

I had to switch to plain physical server (which was my target by then).

...

All seems well according to OpenLDAP manual, but maybe I'm missing something. I start both servers using "slapd -d sync" in order to filter only the sync part.

I created a script to insert 200 employees in my directory, to force the servers to sync up. When I start both servers and insert the employees in machine 1, I get this too much lines with this sample: slap_queue_csn: queing 0xb6851b4a

20090623082909.571025Z#000000#000#000000

...

slap_graduate_commit_csn: removing 0xb82165c8 20090623082909.571025Z#000000#000#000000

So the CSN is being created (I think), but there's no sync to machine 2, which makes me frustated. . .

I'd appreciate any comments you can have, especially if you already have this kind of setup running.

Best regards, Fernando

Hope it can help. Sincerely yours, Mathieu MILLET.