Solaris x86 11/06 , fragmentation & I/O problem, HP DL380 g5, Help !! Thank you - Solaris
This is a discussion on Solaris x86 11/06 , fragmentation & I/O problem, HP DL380 g5, Help !! Thank you - Solaris ; Dear all, Allthough, we successfully installed solaris 10 x86 ver 11/06 , on a HP 380 g5 with p400 smartarray card and SAS 72Gb disks. The problem is that from the beginning fsck displays a file fragmentation much higher than ...
![]() |
| | LinkBack | Tools |
|
#1
| |||
| |||
| Allthough, we successfully installed solaris 10 x86 ver 11/06 , on a HP 380 g5 with p400 smartarray card and SAS 72Gb disks. The problem is that from the beginning fsck displays a file fragmentation much higher than usual ( 1.8 % and up ), doing Oracle import is very slow ( much more slower than Solaris 9 + DL 380g3 ) and the worse is that our database after few days become corrupted. We try to find some clue about what's going on without success. Do you also have the same fragmentation and corruption issue with solaris 10x86 ?? Any advice ??? Thanks Steven |
|
#2
| |||
| |||
| On Jun 29, 2:04 pm, and...@cucumber.demon.co.uk (Andrew Gabriel) wrote: > In article <1183087822.982005.276...@i38g2000prf.googlegroups. com>, > fdouts...@gmail.com writes: > > > Dear all, > > > Allthough, we successfully installed solaris 10 x86 ver 11/06 , on a > > HP 380 g5 with p400 smartarray card and SAS 72Gb disks. > > The problem is that from the beginning fsck displays a file > > fragmentation much higher than usual ( 1.8 % and up ), doing Oracle > > import is very slow ( much more slower than Solaris 9 + DL 380g3 ) and > > the worse is that our database after few days become corrupted. We try > > to find some clue about what's going on without success. Do you also > > have the same fragmentation and corruption issue with solaris 10x86 ?? > > The fragmentation reported by fsck is an absolutely normal feature > of ufs. It is an indication of the number of files which are sharing > a fragmented ufs block in order to avoid wasting a whole ufs block > when they don't need that much disk space. It does not have anything > to do with fragmented allocation layout. > > When/why is fcsk being run? > > -- > Andrew Gabriel > [email address is not usable -- followup in the newsgroup] Hi andrew, Thanks a lot for your update, the point is that even we applied latest patch from HP and SUN ( we have 3 servers in the same bad condition ), the IO are very slow ( compare to solaris9 + dl380 g3 ) and day by day hard disks conditions are deteriorating ( our database is corrupted with bad blocks every 2/3 days, corrupted tables seems the one with the largest volume of Data and the largest IO ). I am quite puzzled because i cannot find a lot of post about these issues and i guess i am not the only one using HP G5 with solaris 10 ! ( we did the default installation ). We tried to create a file partition using ZFS , IO seems faster ( but not exceptionnal ), so i am very wondering that an incompatibilty issue still exists between HP and Solaris 10 , or pehaps it is the SAS hard disk format that is not well managed by solaris....still very very wondering what's going on ! :-) Steven |
|
#3
| |||
| |||
| fdoutside@gmail.com wrote: > Thanks a lot for your update, the point is that even we applied latest > patch from HP and SUN ( we have 3 servers in the same bad condition ), > the IO are very slow ( compare to solaris9 + dl380 g3 ) and day by What does iostat tell you? Is Oracle configured to do use directio? Have you run any benchmarks using a tool such as Bonnie++? I suspect he wanted to know when you're running fsck in order to establish whether or not you're running fsck against an in-use filesystem, which might account for data corruption. -- Brandon Hume - hume -> BOFH.Ca, http://WWW.BOFH.Ca/ |
|
#4
| |||
| |||
| I had a problem quite similar to yours. Please mount the file system (ufs) without caching. I had similar issues with EMC storages, they do not work good with FS cahcing enabled, Maybe this is also the case here. Hope it helps. Kind regards. On Jun 29, 6:30 am, fdouts...@gmail.com wrote: > Dear all, > > Allthough, we successfully installed solaris 10 x86 ver 11/06 , on a > HP 380 g5 with p400 smartarray card and SAS 72Gb disks. > The problem is that from the beginning fsck displays a file > fragmentation much higher than usual ( 1.8 % and up ), doing Oracle > import is very slow ( much more slower than Solaris 9 + DL 380g3 ) and > the worse is that our database after few days become corrupted. We try > to find some clue about what's going on without success. Do you also > have the same fragmentation and corruption issue with solaris 10x86 ?? > Any advice ??? > > Thanks > > Steven |
|
#5
| |||
| |||
| On Jun 30, 7:28 am, pavelj > I had a problem quite similar to yours. > Please mount the file system (ufs) without caching. I had similar > issues with EMC storages, they do not work good with FS cahcing > enabled, Maybe this is also the case here. > Hope it helps. > Kind regards. > > On Jun 29, 6:30 am, fdouts...@gmail.com wrote: > > > Dear all, > > > Allthough, we successfully installed solaris 10 x86 ver 11/06 , on a > > HP 380 g5 with p400 smartarray card and SAS 72Gb disks. > > The problem is that from the beginning fsck displays a file > > fragmentation much higher than usual ( 1.8 % and up ), doing Oracle > > import is very slow ( much more slower than Solaris 9 + DL 380g3 ) and > > the worse is that our database after few days become corrupted. We try > > to find some clue about what's going on without success. Do you also > > have the same fragmentation and corruption issue with solaris 10x86 ?? > > Any advice ??? > > > Thanks > > > Steven Fragmentation can occur on the datablocks in the oracle database. Oracle has it's own tools for dealing with this. The fragmentation on oracle is totally separate from the OS level. Please get a good oracle dba or read the oracle documentation on your version db to see how to use their tools. Solaris is also slow compared to linux for performance. Version 10 is supposed to be better, but we gave up on solaris due to our database vendor (progress) no longer supporting their database on solaris x86 (this was after sun declared that they would NOT support x86 a few years ago). Sun hurt their customers a lot by doing that. Good luck, Oskar |
|
#6
| |||
| |||
| On 2 juil, 02:56, "pheoni...@gmail.com" > On Jun 30, 7:28 am, pavelj > > > > > > > I had a problem quite similar to yours. > > Please mount the file system (ufs) without caching. I had similar > > issues with EMC storages, they do not work good with FS cahcing > > enabled, Maybe this is also the case here. > > Hope it helps. > > Kind regards. > > > On Jun 29, 6:30 am, fdouts...@gmail.com wrote: > > > > Dear all, > > > > Allthough, we successfully installed solaris 10 x86 ver 11/06 , on a > > > HP 380 g5 with p400 smartarray card and SAS 72Gb disks. > > > The problem is that from the beginning fsck displays a file > > > fragmentation much higher than usual ( 1.8 % and up ), doing Oracle > > > import is very slow ( much more slower than Solaris 9 + DL 380g3 ) and > > > the worse is that our database after few days become corrupted. We try > > > to find some clue about what's going on without success. Do you also > > > have the same fragmentation and corruption issue with solaris 10x86 ?? > > > Any advice ??? > > > > Thanks > > > > Steven > > Fragmentation can occur on the datablocks in the oracle database. > Oracle has it's own tools for dealing with this. The fragmentation on > oracle is totally separate from the OS level. Please get a good > oracle dba or read the oracle documentation on your version db to see > how to use their tools. Solaris is also slow compared to linux for > performance. Version 10 is supposed to be better, but we gave up on > solaris due to our database vendor (progress) no longer supporting > their database on solaris x86 (this was after sun declared that they > would NOT support x86 a few years ago). Sun hurt their customers a > lot by doing that. > > Good luck, > > Oskar- Masquer le texte des messages précédents - > > - Afficher le texte des messages précédents - Dear All, Thanks a lot for all your post , i am just back from few days break. Yes, we indeed fsck on unmounted disks, the purpose of fsck was to try to find a clue about our frag' problem. I will checl you suggestion on caching and other Oracle tuning if any. Keep u guys update. Thanks again to the community, it rocks !!! |
|
#7
| |||
| |||
| Dear All, We switched from UFS file system to ZFS and things seem to improve a bit. So far result of our tests : - DL380 G5 / Smart array P400 or P800 + SAS disks = Not OK , ( ZFS or UFS ) - DL380 G5 / Smart array P600 + SAS disks = Not OK with UFS , so far seems OK with ZFS I guess we are encountering so far some incompatibilty problems with smart array P400 / P800 and solaris 10 x86 and also perhaos some issues with SAS disk on solaris 10. What still amaze me , is the lack of information about this on the web ( i cannot believe we are the only one using this particular config ! ) Thanks Steven |
|
#8
| |||
| |||
| fdoutside@gmail.com wrote: > We switched from UFS file system to ZFS and things seem to improve a > bit. So far result of our tests : I'd still be a bit worried... just because ZFS might be correcting/rebuilding corruption on the fly doesn't mean it isn't happening. I'd also be nervous about running Oracle on ZFS, which is just a bit too new (and I've heard of plenty of performance problems with the combo, as well). I'd really recommend allocating a chunk of disk space, and running some filesystem performance tools like Bonnie++, and perhaps the analysis and verification options in the format tool. > What still amaze me , is the lack of information about this on the web > ( i cannot believe we are the only one using this particular > config ! ) Solaris x86 is still recovering from previous Sun neglect. You might also be in the odd middle ground between "cheap" (Linux on x86) and "serious" (Solaris on SPARC). -- Brandon Hume - hume -> BOFH.Ca, http://WWW.BOFH.Ca/ |
|
#9
| |||
| |||
| On Jul 4, 8:43 am, hume.spamfil...@bofh.ca wrote: > fdouts...@gmail.com wrote: > > We switched from UFS file system to ZFS and things seem to improve a > > bit. So far result of our tests : > > I'd still be a bit worried... just because ZFS might be correcting/rebuilding > corruption on the fly doesn't mean it isn't happening. I'd also be nervous > about running Oracle on ZFS, which is just a bit too new (and I've heard of > plenty of performance problems with the combo, as well). > > I'd really recommend allocating a chunk of disk space, and running some > filesystem performance tools like Bonnie++, and perhaps the analysis and > verification options in the format tool. > > > What still amaze me , is the lack of information about this on the web > > ( i cannot believe we are the only one using this particular > > config ! ) > > Solaris x86 is still recovering from previous Sun neglect. You might also > be in the odd middle ground between "cheap" (Linux on x86) and "serious" > (Solaris on SPARC). > > -- > Brandon Hume - hume -> BOFH.Ca,http://WWW.BOFH.Ca/ Like I said before, we got out of solaris due to lack of support (progress database) and poor performance on x86. Linux is far superior in performance on x86 boxes. since you have HP servers, Suse or redhat work much better on HP servers. CentOS is a clone of redhat enterprise that is free. We use suse a lot (great admin tools, far better than solaris or redhat), suse runs awesome on HP servers and has excellent driver support. Suse also has excellent support from Novell. Speaking about vendor support, did you get the support contract from Sun? This would be a great time to call them. They also have very good support. It looks like your company is spending serious $$$ on this database, why risk running it on a flaky system? Solaris x86 has always had poor support when it comes to hardware/drivers. In regards to admin tools, Suse and Apple are far ahead of anyone else. Both make it easy to work on unix/linux. When was the last time you saw that? Good luck, Oskar |
|
#10
| |||
| |||
| On Wed, 04 Jul 2007 19:33:36 -0700, pheonix1t@gmail.com > > Like I said before, we got out of solaris due to lack of support > (progress database) and poor performance on x86. Linux is far > superior in performance on x86 boxes. since you have HP servers, Suse Please post some proof of this superior performance. Until you do, I don't believe a word of this claim. From what I've read, Solaris 10 performs well on multicore x64. I can well imagine that on old 32bit hardware Linux performs better. As for HP, my current experience says 'don't touch them with a bargepole'. > It looks like your company is spending serious $$$ on this database, > why risk running it on a flaky system? Solaris x86 has always had > poor support when it comes to hardware/drivers. I've never had any problems with my hardware. > In regards to admin > tools, Suse and Apple are far ahead of anyone else. Both make it easy > to work on unix/linux. When was the last time you saw that? Eh? I'm writing this on a Mac, and other than the dialog boxes being a bit prettier, it was no easier to set up. A bientot Paul |
|
#11
| |||
| |||
| I've just had my second glass of wine. Iv'e been an oracle DBA for 12 years. Look you are all talking rubbish. Solaris performance is really hot and linux is horrid to administer. But the orginal post is about an import. the import of data is not a real test. of performance if your import is slow look at the commit rate and buffers size and the striping of disks( spread the I/O load). the operating system really shouldn't be getting in the way. it should be a load to hardware thing. so you should be looking at the io rates. remembering that the commit has to happen and write a header to all datafiles. so increase the number of rows processed before the commit. ( its a classic problem) Again HP gear is beautfully built as is sun just get the db processing right. Cheers Hope That helps Paul Floyd wrote: > On Wed, 04 Jul 2007 19:33:36 -0700, pheonix1t@gmail.com > > >> Like I said before, we got out of solaris due to lack of support >> (progress database) and poor performance on x86. Linux is far >> superior in performance on x86 boxes. since you have HP servers, Suse >> > > Please post some proof of this superior performance. Until you do, I > don't believe a word of this claim. > > From what I've read, Solaris 10 performs well on multicore x64. I can > well imagine that on old 32bit hardware Linux performs better. > > As for HP, my current experience says 'don't touch them with a > bargepole'. > > >> It looks like your company is spending serious $$$ on this database, >> why risk running it on a flaky system? Solaris x86 has always had >> poor support when it comes to hardware/drivers. >> > > I've never had any problems with my hardware. > > >> In regards to admin >> tools, Suse and Apple are far ahead of anyone else. Both make it easy >> to work on unix/linux. When was the last time you saw that? >> > > Eh? I'm writing this on a Mac, and other than the dialog boxes being a > bit prettier, it was no easier to set up. > > A bientot > Paul > > |
|
#12
| |||
| |||
| On Thu, 05 Jul 2007 20:16:47 GMT, Mike Duffy > Again HP gear is beautfully built as is sun just get the db processing > right. It may be so, but sadly here their service is so bad that they seem to be incapable of delivering what I ordered. Honestly, I wish I'd ordered a Dell. A bientot Paul |
|
#13
| |||
| |||
| On Jul 5, 3:31 pm, Paul Floyd > On Thu, 05 Jul 2007 20:16:47 GMT, Mike Duffy > > Again HP gear is beautfully built as is sun just get the db processing > > right. > > It may be so, but sadly here their service is so bad that they seem to > be incapable of delivering what I ordered. > > Honestly, I wish I'd ordered a Dell. > > A bientot > Paul Mr DBA (previous post), thank you. If you look at MY previous posts, I suggested that the poster talk with or get an oracle dba to look at this. As for Paul, I'd love to post performance tests but public ones are rare to come by since vendors make it illegal to make performance tests public without their consent. I've worked in IT for 16 years (hpux was my first unix at work, then solaris). If you've never seen suse, please don't say anything about it, it only make you look foolish. The admin tools from suse are much better than just about anything else on the market. Sun hasn't improved the management tools in YEARS. When I mention apple I'm talking about their servers. Again, if you've never worked on them please don't dismiss what I'm saying. solaris on x86 has always been poor in performance. It got better for version 10, but not as fast as linux. I've run suse and solaris 10 on same hardware, suse is faster by a long shot on simple things (users doing queries on database, samba file transfers, apache web sites working faster). If it does this better, the hard stuff would only be better too (keep in mind, this is on same hardware). I like sun, but their support on x86 has not been consistent and it shows. The place I work at just bought $15k HP server that was a better server than anything sun had for same price. $15k isn't enterprise class server, but for small/medium business this is a hefty sum. i did my research. We also (due to progress) had to look at other choices. Dell has been a bad experience everywhere I've had to work with it. One place (in 2004) bough 650 workstations and 12 servers. All 12 servers had recall due to main boards burning up and smoking heavily! Imagine sites across country starting to call within 6 weeks saying all the servers are burning! workstations weren't much better, in less than 8 months more than half of the workstations had cdroms, disks or mainboards that went bad. I was amazed by it. But they are the cheapest....this showed me you get what you pay for. Anyway, mr DBA is the one who we should be paying attention to! cheers mr DBA ![]() |
|
#14
| |||
| |||
| On Thu, 05 Jul 2007 19:03:33 -0700, pheonix1t@gmail.com > On Jul 5, 3:31 pm, Paul Floyd >> On Thu, 05 Jul 2007 20:16:47 GMT, Mike Duffy >> > Again HP gear is beautfully built as is sun just get the db processing >> > right. >> >> It may be so, but sadly here their service is so bad that they seem to >> be incapable of delivering what I ordered. >> >> Honestly, I wish I'd ordered a Dell. > > Mr DBA (previous post), thank you. If you look at MY previous posts, > I suggested that the poster talk with or get an oracle dba to look at > this. > As for Paul, I'd love to post performance tests but public ones are > rare to come by since vendors make it illegal to make performance > tests public without their consent. > > I've worked in IT for 16 years (hpux was my first unix at work, then > solaris). If you've never seen suse, please don't say anything about > it, it only make you look foolish. The admin tools from suse are much I have used SUSE, though not recently. I didn't find it significantly easier to install than Solaris. The trickiest thing on Solaris, for me, was getting a networked PCL printer to work. That could have been easier, I must admit. > better than just about anything else on the market. Sun hasn't > improved the management tools in YEARS. There were significant changes in Solaris 10. In particular to the services. > When I mention apple I'm talking about their servers. Again, if > you've never worked on them please don't dismiss what I'm saying. Never used an Apple server, not even seen one. I have a Mac portable. Don't imagine that we'll ever have anything from Apple at work, ever. > solaris on x86 has always been poor in performance. It got better for > version 10, but not as fast as linux. I've run suse and solaris 10 on > same hardware, suse is faster by a long shot on simple things (users > doing queries on database, samba file transfers, apache web sites > working faster). If it does this better, the hard stuff would only be > better too (keep in mind, this is on same hardware). I'm not involved with databases, I'm a software developer. Perhaps Linux is handicapped by the slowness of GCC and GNU binutils, but on the same hardware, compiling/linking on Solaris is generally faster than on Linux. [Dell horror stories] We have perhaps 50 Dell workstations at work. Nothing much to report about them. My beef with HP is that they seem to choose the cheapest subcontractors for their distribution, and they don't do a very good job. But they finally seem to be getting their act together. A bientot Paul |
|
#15
| |||
| |||
| I couldn't help digging this one up. gcc uses /tmp which is ram on Solaris, disk in Linux. Quote:
|
![]() |
« Previous Thread
|
Next Thread »
| Tools | |
| |
Similar Threads | ||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| Installing SCO 5.0.7 on HP DL380 G5 | unix | SCO | 18 | 03-15-2008 12:07 AM |
| GBit speed problem - TG3 on HP DL380 G3 | unix | Networking | 4 | 01-20-2008 04:09 AM |
| DL380 G4 | unix | Hewlett Packard | 2 | 10-04-2007 11:57 PM |
| Strange network problem, collisions and fragmentation | unix | Network | 3 | 10-03-2007 06:09 AM |
| A tricky problem concerned with fragmentation and Logical link Layer service. | unix | TCP-IP | 2 | 10-02-2007 09:54 PM |
All times are GMT. The time now is 08:24 AM.






