Maser Posted June 19, 2002 Report Share Posted June 19, 2002 This has become 100% reproducable for me... I have 4 machines running Retro 5.0.205 on G4/533s with 384M RAM running OSX 10.1.5 backing up to a DLT drive over Adaptec 29160 cards (current drivers). No other programs are running. There is no "classic" on these machines. All 4 of them are set to run in "backup server" mode 24x7. All 4 of them are having the "LaunchCFMA" process (which is Retrospect) suck RAM: This is what "top" says after I restart the computer and launch Retrospect (the "starting point"): Processes: 35 total, 2 running, 33 sleeping... 82 threads 08:41:53 Load Avg: 0.59, 0.25, 0.15 CPU usage: 8.4% user, 4.7% sys, 86.9% idle SharedLibs: num = 78, resident = 19.3M code, 1.36M data, 5.02M LinkEdit MemRegions: num = 1317, resident = 17.4M + 5.45M private, 22.1M shared PhysMem: 31.4M wired, 42.6M active, 51.5M inactive, 126M used, 258M free VM: 856M + 41.9M 2458(1) pageins, 0(0) pageouts PID COMMAND %CPU TIME #TH #PRTS #MREGS RPRVT RSHRD RSIZE VSIZE 277 top 4.5% 0:00.19 1 14 15 192K 292K+ 448K+ 1.62M 272 tcsh 0.0% 0:00.08 1 24 15 440K 620K 912K 5.72M 271 TextEdit 0.0% 0:00.40 1 64 68 1008K 4.87M 3.64M 60.5M 270 Terminal 2.7% 0:01.08 4 68 99 1.65M+ 4.52M 4.04M+ 62.7M 269 RetroRun 0.0% 0:00.01 1 11 16 152K 1.25M 148K 2.54M 263 LaunchCFMA 0.0% 0:01.80 3 83 82 2.23M 7.39M 5.83M 62.8M This is before I quit the program (program has been running for one week): Processes: 35 total, 2 running, 33 sleeping... 89 threads 08:33:40 Load Avg: 1.21, 0.56, 0.33 CPU usage: 16.5% user, 18.3% sys, 65.2% idl SharedLibs: num = 84, resident = 12.2M code, 940K data, 2.83M LinkEdit MemRegions: num = 6118, resident = 302M + 2.54M private, 15.6M shared PhysMem: 40.3M wired, 226M active, 113M inactive, 379M used, 4.67M free VM: 1.73G + 42.9M 76848041(1) pageins, 6960225(0) pageouts PID COMMAND %CPU TIME #TH #PRTS #MREGS RPRVT RSHRD RSIZE VSIZE 635 LaunchCFMA 8.8% 102 hrs 7 121 3207 292M 5.15M 289M 872M This is after I quit the program: Processes: 34 total, 2 running, 32 sleeping... 81 threads 08:37:53 Load Avg: 0.51, 0.28, 0.18 CPU usage: 5.4% user, 11.7% sys, 82.9% idle SharedLibs: num = 84, resident = 13.2M code, 916K data, 2.77M LinkEdit MemRegions: num = 2888, resident = 9.99M + 2.45M private, 12.8M shared PhysMem: 40.1M wired, 20.7M active, 25.4M inactive, 86.2M used, 298M free VM: 891M + 42.9M 76895311(2) pageins, 6977449(0) pageouts (no LaunchCFMA) running in top... On all 4 of my machines, "LaunchCFMA" creeps up steadily. This doesn't affect the speed of the backups, but if I use Retrospect for any other purpose (ie, add a new client), the "spinning-wheel" comes back. Anybody else seeing anything similar? - Steve Link to comment Share on other sites More sharing options...
Maser Posted June 21, 2002 Author Report Share Posted June 21, 2002 Replying to my own topic... I've gotten 3 e-mail messages from people who said they are having the exact same problem (no other software running, yet LaunchCFMA creeps up in RAM as Retrospect continually runs.) Any comment from Dantz about this? Link to comment Share on other sites More sharing options...
CallMeDave Posted June 21, 2002 Report Share Posted June 21, 2002 Steve, I don't understand your recent post where the LaunchCFMApp process has a different PID (Processor ID) number between the "Starting point" and "after one week" examples. Since PID's don't change on their own, and since my use of Retrospect on multiple machines w/SCSI devices do not show a memory leak like you do, what is happening during that week? Dave Link to comment Share on other sites More sharing options...
Maser Posted June 24, 2002 Author Report Share Posted June 24, 2002 Nothing is happening during the week. Honest. The machine is rebooted, the new set of tapes are put in and it runs all week backing up everything. At the end of the week (when we'd cycle the tapes), that's when we notice how sluggish the non-backup operations of Retrospect are... Link to comment Share on other sites More sharing options...
Maser Posted June 24, 2002 Author Report Share Posted June 24, 2002 Since the problem was happening again since I last rebooted on *Wednesday* and it's showing the same problem since the weekend, I checked it again. Top showed this: 263 LaunchCFMA 9.5% 72:58:10 8 122 3339 298M 4.74M 293M 865M Running "ps -auxww" showed this as the only instance of launchCFMA: USER PID %CPU %MEM VSZ RSS TT STAT TIME COMMAND root 263 8.8 76.4 885916 300432 ?? S 4378:10.97 /System/Library/Frameworks/Carbon.framework/Versions/A/Support/LaunchCFMApp /Applications/Retrospect 5.0/Retrospect/Contents/MacOS/AuthenticateUser.app/Contents/MacOS/../../../Retrospect So, It's pretty clear to me who the culprit is... The question is why is this happening? Link to comment Share on other sites More sharing options...
smithkennedy Posted June 26, 2002 Report Share Posted June 26, 2002 I have been seeing the same problem. I had to add more RAM to my backup server to prevent performance problems. I have seen the "LaunchCFMApp" associated with "RetroRun" take in excess of 300 MB of RAM. It seems to grow when it is matching against previous sets, which I guess Retrospect is loading them all into RAM, but not discarding right away. I noticed a similar problem with Retrospect 4.3, which forced me to increase the memory for the Retrospect application to something like 60MB, or else it would hang the system. I have been blown away at the voracious appetite Retrospect has had for memory in Mac OS X. I hope they will someday do a Mach-O binary, but I don't know if that will solve the problem. My System: G3 400 (B&W), 768MB RAM, Adaptec 29160 (not the 29160N), HP C5713A SureStore 40x6 DAT Autoloader, Mac OS X Server 10.1.4 (with 10.1.5 update installed) Link to comment Share on other sites More sharing options...
IrenaS Posted June 26, 2002 Report Share Posted June 26, 2002 The 29160 card isn't supported I'm afraid. Only the 29160N card is supported at this time. Irena Solomon Dantz Tech Support Link to comment Share on other sites More sharing options...
smithkennedy Posted June 26, 2002 Report Share Posted June 26, 2002 Hmm, that's a bummer. O.K., after checking out the Dantz supported SCSI adapter page: http://www.dantz.com/index.php3?SCREEN=osx_scsi_adapter&sid=UmKWSlPru3nm89za and noticing that my tape drive has a wide interface, and there is a notice that you shouldn't connect a wide device to a narrow card with Retrospect (which was proven by my using a 2930U card, and causing a total hang of the backup server), which card would you recommend to connect this? I have seen stuff in these discussions that make me wary of 39160 cards, too. Dantz seems to like the Atto cards, would an EPCI-UL3S be a card that Dantz has had good luck with? Link to comment Share on other sites More sharing options...
IrenaS Posted June 26, 2002 Report Share Posted June 26, 2002 Any supported card with the same pin configuration as your tape drive will be supported. Other forum users have certainly reported success with ATTO cards; most of the failures reported are due to people using cards that are not supported or narrow cards with wide drives. Link to comment Share on other sites More sharing options...
Peter_Haase Posted June 26, 2002 Report Share Posted June 26, 2002 It does not matter if you use ATTO cards. I switched from Adaptec to ATTO and and still seeing LaunchCFMA gobbling up memory as well as a few other oddities. Dantz seems to be keeping the lid on all their problems with the OS X version of the product. Very frustrating. Link to comment Share on other sites More sharing options...
leechubb Posted June 26, 2002 Report Share Posted June 26, 2002 I'm using the plain old Superdrive in my G4 tower, so there should be no compatibility issue there. I came in on Monday and found that the machine was very sluggish, took a look at the processes and found that Retrospect (under its alias - LaunchCFMApp) had taken up 55% percent of the very large memory in that box! This is ridiculous, and we need a fix SOON. It's there, it's reproducible, and it directly affects whether I can use Retrospect as a viable backup solution. I can't right now, because it's interfering badly with the other services on that box. It would be hard to overstate how frustrated I am with Dantz right now. Between this and the "clicking defer doesn't defer" bug, I spend half my time just trying to keep Retrospect from getting in the way of the actual work that's supposed to get done around here. Build 205 was released ages ago now and there has been no new progress on getting serious bugs fixed. When will fixes be released? Link to comment Share on other sites More sharing options...
Maser Posted June 26, 2002 Author Report Share Posted June 26, 2002 Agreed. While, sure, it could be my 29160 card (which I'm having *zero* problems with), that wouldn't explain the RAM issue, would it? The other users here at the UofM had other SCSI cards and saw the problem, too. I'm needing to reboot every 4 days or so as I can't add clients (much less modify anything) without the spinning wheel whenver LaunchCFMA goes above my hard 384M RAM. I could probably get around this by putting 1G of RAM in each backup machine (as I've never seen LaunchCFMA go above 840) -- which is what one department admin did here -- but I'm not looking to address a software issue with a hardware solution if it's not necessary. There's a memory leak somewhere. I'm open to suggestions to help debug it. Link to comment Share on other sites More sharing options...
IrenaS Posted June 26, 2002 Report Share Posted June 26, 2002 We are aware of this and we are certainly looking into it. When more information is available, it will be posted to the forum. Thanks, Irena Solomon Dantz Tech Support Link to comment Share on other sites More sharing options...
CallMeDave Posted June 26, 2002 Report Share Posted June 26, 2002 In reply to: There's a memory leak somewhere. I'm open to suggestions to help debug it. A good way to start would be to describe your setup in detail. It's not happening in our office, so there must be something that brings it on that we're not doing but that you are. - For example, how many sessions are in your Backup Set? - How many Snapshots? - How many clients are being polled/scanned by Backup Server? - Does it behave this way when you Recycle the Backup Set? - Do you have any non-default settings in your Backup Server Script (such as polling time)? Dave Link to comment Share on other sites More sharing options...
Maser Posted June 27, 2002 Author Report Share Posted June 27, 2002 - For example, how many sessions are in your Backup Set? doesn't matter. This will happen after starting a new set as well. It can be as low as 60 (after one week to a new set of tapes) to as high as 400+ - How many Snapshots? Again, doesn't matter. I don't retain more than the current snapshot (though each is "saved", none but the current are loaded...) - How many clients are being polled/scanned by Backup Server? At a minimum, 65 on one backup server machine. At a maximum, 120 on another. (The other two are in the upper 70s.) - Does it behave this way when you Recycle the Backup Set? Yes. - Do you have any non-default settings in your Backup Server Script (such as polling time)? non-defaults: for most of my scripts, I backup once every 7 days. On two of the systems, I run a 2nd script that backs up a small number of machines (1 and 4) every day. I also do *not* do verification (there are reasons for this) the only other non-default is to retry after 99 minutes. I'd set that larger if I could... Link to comment Share on other sites More sharing options...
Recommended Posts
Archived
This topic is now archived and is closed to further replies.