Major Performance Issues after Upgrade from 5.0.16 to 6.0.6

[Aragorn]

Hi there,

I've just upgraded our production zimbra server to 6.0.6 from 5.0.16. The upgrade went successfully, nothing untoward happened and the server started back up nicely.

Its sending and receiving mails as normal, and IMAP/POP access seems fine.

The problem is the web interface. It now seems to be taking a very long time to log in, and once logged in, certain activities such as viewing the calendars seem to take a long time to appear. The calander will open and show as blank, but after a minute or two the data will appear.

It seems, at least in part, related in some way to how full the users account is, my admin account for instance has no mail in it, and loads instantly.

The server itself doesnt appear to be under any sort of major load when this occurs, perhaps 5-10% cpu but nothing more. Theres over a GB of RAM free too.

I've had a look thru some of the log files and cant see anything of note, so i'd really like some pointers where to look next?

Cheers
Kevin

[uxbod]

Welcome to the forums

How powerful are the client machines ? Lazy loading is used so it could be that the Ajax is taking a while to process.

[Aragorn]

Thanks!

Various specs i guess, most lowly machine in the office is probably late P4 era, but they're all stacked up with ram as they get used for a lot of CAD work.

I've tried using the standard HTML mode at login time, and that causes every page load to take a long time.

I've just done a test on my laptop with one of the guys accounts, upon clicking "view mail" in the admin panel, it pops a new window which states Transferring data from zimbra... in the status bar. CPU utilisation was nil at this point. After approx 90 seconds, zimbras Loading... page appeared, CPU shot up to 50% (its a dual core) and after a second or so the inbox loaded.

Any ideas where i should be looking for error messages?

[LMStone]

Slow loading times like this often encourage us to look at some networking fundamentals, like switches, the cable plant, firewall config etc.

The thinking is that: If the Zimbra server is not the bottleneck and the workstations are powerful enough then what's between the server and the workstations bears some investigation.

We had a new medical practice client at first complain their practice management system was dog slow (it was). They were in a new building, their servers were barely loaded, the workstations were all new Dells, and their previous provider had convinced them to buy high-end Cisco GB switches. What we found was that the Cisco switches didn't auto-neg reliably with the NICs in the Dells, and pretty much everything was talking at 10HD instead 1000FD. Once we fixed that, things flew.

I'm not saying that's your issue here, only that if you are stumped elsewhere it wouldn't hurt to look at the networking side of things.

Hope that helps,
Mark

[uxbod]

Oh I so wish people would fix the neg speed on ports! How many devices are now 10/HD; I could understand that they were not 1000/FD and instead 100/FD.

[Aragorn]

The zimbra server is on the internet, and we're accessing it over DSL lines.

The point is before the update it was perfect, now its not!

[LMStone]

Originally Posted by Aragorn

The zimbra server is on the internet, and we're accessing it over DSL lines.

The point is before the update it was perfect, now its not!

Many have noted here in the forums that ZCS6 can be "slower" than ZCS5 after an upgrade; ZCS6 has more features and we find is more "chatty" than ZCS5.

Assuming your ZCS server is in a proper data center where bandwidth, latency and jitter are not issues, I would ask if you have done any testing on your DSL line to find out what percentage of your bandwidth you are using, and what sorts of latency and jitter you are seeing on the line as well?

Hope that helps,
Mark

[LMStone]

Originally Posted by uxbod

Oh I so wish people would fix the neg speed on ports! How many devices are now 10/HD; I could understand that they were not 1000/FD and instead 100/FD.

It's one of the major reasons we like HP ProCurve switches over Cisco; HP does much, much better autoneg than Cisco. Lifetime warranty, trade-in and free access to firmware updates with no annual SmartNet costs either.

All the best,
Mark

[Hivos]

Same issue here after an upgrade from 5.0.13 --> 5.0.23 --> 6.0.6.

At 5.0.23 all was still fine, but 6.0.6 is much slower. Not sure about the cause though, if it's server, connection (20Mbit DSL) or perhaps browser (FF3.6.3) related. Will look into this.

[shan]

We just upgraded from 5.0.21 to 6.0.6 We are having a bad day - slowness cross-board. I have filed a support case.