Community
Participate
Working Groups
"Ian Graham" <ian@kelman.com> wrote in message news:3D9DE36B.5010803@kelman.com... > I've recently started using Eclipse on a SunBlade 100 running Solaris 8. > The performance is appalling! It takes about 4 seconds to switch > between standard Resource and Java perspectives within the same window > with no editors open. > > On my six-year old PC at home, running NT with only 256M it takes about > a second. On my coworker's snappy PC running Windows 2000 it takes > maybe 0.2 seconds, if that. > > What gives? Is it X, is it Motif, is it SWT/Motif? We've run some > simple benchmarks between the fast PC and the SunBlade, and they > indicate a roughly 4 times speed difference running non-GUI stuff - that > doesn't account for anywhere near the difference I'm seeing in the speed > of Eclipse between the two platforms, so it's not a hardware issue. > > Assuming I don't have a bad build of eclipse full of debug stuff or > something, this really impacts what kind of machine I want to run on. > > I may be able to get a new machine here, and I'm trying to figure out > what to ask for. There's a lot of convenience here on our local net for > me to be using a Unix box, but I'm worried that only the Windows > implementation of SWT will give me usable performance. > > Please tell me I've got a bad build of Eclipse for Solaris! (2.0.1 build > 200208291828 >
GG to investigate. Marking this as a P2. Adding SSQ and myself to the CC because of our changes to menu code. Could it be the "mnemonics slowdown" that was supposed to have gone away?
It's not the mnemonics/accelerators issue since its the 2.0.1 release.
Not so sure. We put back the original suspect mnemonic code. Try deleting it.
In the October 1st build the perspective scenario that's particularly slow is the creation of new perspectives (eg.- start a new workspace, Window -> Open Perspective -> Java). Switching between perspectives that are already open is reasonably quick. I profiled the creation of a new Java perspective on Solaris. Considering the methods with % > 1.0%, Solaris spends > 65% of this time doing class loading (dynamic retrieval of the jdt components). On Linux this number is ~30%. The generated hot spot lists are attached. So I don't think there's much for swt to do here. I'll look at some other scenarios while I'm at it.
Created attachment 2195 [details] solaris hot spots: new Java perspective
Created attachment 2196 [details] linux hot spots: new Java perspective
I guess I should clarify that in my initial posting I was talking about perspectives that ARE already open. When you say that "Switching between perspectives that are already open is reasonably quick", what do you mean by "reasonably quick"? For me, on this SunBlade 100, it really is taking over four seconds between already open Java Browsing and Java perspectives. And that slow speed can be found in everything I do in Eclipse, where even just switching between java editors takes over a second to update the Outline. Ian
Changing between perspectives here takes < 3 seconds on Solaris, which isn't as fast as win32 but isn't disruptive either. So if the question is "without considering class loading why is Eclipse on win32 faster?", it's because Motif is slower in comparison. Eclipse on win32 is also faster than Eclipse on gtk and photon. I'm still going to do more profiling, but my initial look yesterday gave me confidence that we are not doing anything overly inefficient to make Eclipse on Solaris (Motif) run slower than it could given the requirements of the layers above us.
Changing to P3 since performance analysis will extend beyond the 2.1 release, so this report cannot be finalized.
Adding my name to the cc list as we are now tracking performance issues more closely. Please remove the performance keyword if this is not a performance bug.
Ian, this problem report is really old. How are things in the 3.0 stream ... wince ...
Hi Steve, not long after all this I got a Windows PC and stopped worrying about it. For my newest project I'm using RCP and no longer targeting Solaris - our company is essentially moving towards Intel/Linux desktops. Although I still don't know why using Eclipse was _so_ slow on a SunBlade 100, the hardware was still inadequate for our Java apps. On a modern Intel/Linux box Eclipse and SWT seem pretty snappy, so I've never looked back at Solaris. So I guess my answer is that I don't know how things are now in the 3.x stream :)
Thanks for the information. GG to determine if this is still an issue.
This bug has some neat profiling information, but is really too old and vague to define any work to be done. Comment #8 I think summarizes the results of this bug best. I don't see the use in keeping this bug open. Please file new bugs if you are seeing specific performance issues on this platform.
