We have had quite a few performance problems and a lot of growing number of sde connections. This session might enlighten me.
A number of things has been done in recent releases to enhance performance; to mention two: shared cursors and fewer round trips to the server.
A small survey on how many uses Application Server Connection vs. Direct Connect? Seems to be fifty fifty. For better performance you would choose Direct Connect for these reasons:
- Direct Connect is 2 tier ( vs. 3 tier for ASC)
- Procesing on the client instead of the server (less use of heap memory on the server)
'How many have experinced long connect times?' we were asked and quite a few could say YES. So we got an explanation on what happens: in order to connect there is created a named list of objects (table names) if you have a lot of tables you see what happens. Then these list has to be filtered ... you see the problem. What you should do:
- Remove unused tables from your production database
- Use Roles! By using roles you can narrow down access to tables creating lesser numbers of objects.
After this lecture we were taken a tour on the various sde tools. First one being sdeconfig. Here the important news is that you set RASTERBUFSIZE as low as possible. That differs from what was true prior to 9.2. Reason is that they have optimized code to use threading and allocating huge memory for each is just not good.
Turning on log files and error logging is ONLY for debugging purposes. Do remember to turn it off again!
TCPKeepAlive makes sure that created connections are kept open even though not used. This didn't work in 9.2 and is not fixed until SP5!
SDEIntercept and SDETrace are good tools for monitoring what net work activity is on. Do switch off after use.
Info: SQLServer2008 not yet certified.
Large features should be split up in smaller ones and the use of many layers in your MXD files should be avoided if possible. A good idea is to merge features and use subtyping instead.
Spatial index is automated in 9.2. Recalculation possible with ArcCatalogue based on real feature sampling. Note that deleting features create skewed index.
Loading of raster has been improved with 9.3 (uses all CPUs on the box).
Use client map caching locally on your desktop. Turn on caching at a certain zoom level and with layers you want to see turned on. Creation can take a little while but afterwards you can pan around without getting data from the server.
Versioned layers: Reconcile and post AND compress. There are right now no tools to tell when to do this, but there are plans to include this in future releases.
That was it.
/Sik
Ingen kommentarer:
Send en kommentar