Re: another error with md malloc based fs
Hi Oliver, Hi @list, yes that's exatly what i have in my mind, after the explainings from Chuck. thanks very much cheers michael 2007/3/23, Oliver Fromme <[EMAIL PROTECTED]>: Michael Schuh wrote: > Chuck Swiger wrote: > > Michael Schuh wrote: > > > i can't understand how malloc can eat all available > > > memory, i have 2Gigs of it ;-) > > > so it seems to me i know what i doing, if > > > i have 1,6 Gigs free Memory, and i say ok get me 750Megs from > > > my 1,6 Gigs of free Memory, what was faulty on that > > > > The two choices involve a swap-based RAMdisk, which can use all of > > the available physical RAM it needs to, since this memory is > > swappable, or a kernel-memory-based RAMdisk, which uses wired-down > > memory from within the kernel. > > Ok, so that is it harder to understand for me in the first time, > if i understand it now right the swap or file based memory backend > is also in the ram, but it get another way managed from kernels vm. > the malloc based ramdisk get's not reallly managed by the kvm, > but it underlays under the paging and swapping, and also > ba the "thread-killer" there shot's thread down it it get's to much > work for the system > > i hope i understand this right, it is important for me... Let me try to explain it with a little different words. The following is a bit simplified, but it should give you an idea about the advantages and disadvantages of each md disk type. First of all, _both_ "malloc" and "swap"-backed md disks are RAM disks. You don't have to worry that a swap-backed md disk will be created on your swap partiton. It's not. It will reside in RAM, just like a "malloc" md disk. However, the difference is that a malloc md disk uses memory from the KVM area (kernel virtual memory), which is hardwired into the RAM. It cannot be paged, and it is taken from the kernel's own memory pool, which is usually much smaller than your total available RAM. It's taken from the same pool of RAM that's used for network buffers, driver data and similar things. On the other hand, "swap-backed" md disks use regular memory, so to speak, just like a normal user process. That also means that they are pageable, which means that they can be paged to swap if necessary (if the systems runs low on RAM). That's what "swap-backed" means. Because of that, they don't have a size limit (well, they shouldn't be bigger than your RAM + swap, of course). If there is enough RAM, then they will stay completely in RAM and will _not_ touch your swap at all. For typical RAM disks, such as for /tmp, you should use a swap-backed md disk (it should be the default). The "malloc" type should be used only for rather small, special uses. I hope it's now a little clearer. Best regards Oliver -- Oliver Fromme, secnetix GmbH & Co. KG, Marktplatz 29, 85567 Grafing b. M. Handelsregister: Registergericht Muenchen, HRA 74606, Geschäftsfuehrung: secnetix Verwaltungsgesellsch. mbH, Handelsregister: Registergericht Mün- chen, HRB 125758, Geschäftsführer: Maik Bachmann, Olaf Erb, Ralf Gebhart FreeBSD-Dienstleistungen, -Produkte und mehr: http://www.secnetix.de/bsd "[...] one observation we can make here is that Python makes an excellent pseudocoding language, with the wonderful attribute that it can actually be executed." -- Bruce Eckel -- === michael-schuh.net === Michael Schuh Preußenstr. 13 66111 Saarbrücken phone: 0681/8319664 mobil: 0177/9738644 @: [EMAIL PROTECTED] === Ust-ID: DE251072318 === ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Re: another error with md malloc based fs
Kevin Oberman wrote: > I thought the warning in the man page was adequate (until 7.0 changes > the default to swap backed), but, at least in Michael's case, I guess I > was wrong. I guess the man page need to somehow make it clear that the > memory used by malloc backed mds is a very limited resource. I think it's not necessarily only the language barrier, but the fact that not everyone is familiar with technical terms and details. I guess that many people don't know what "swap-backed" means exactly. I also think that the name "malloc" is badly chosen in that context ... even a (userland) programmer might confuse it with malloc(3), which is quite a different thing. Best regards Oliver -- Oliver Fromme, secnetix GmbH & Co. KG, Marktplatz 29, 85567 Grafing b. M. Handelsregister: Registergericht Muenchen, HRA 74606, Geschäftsfuehrung: secnetix Verwaltungsgesellsch. mbH, Handelsregister: Registergericht Mün- chen, HRB 125758, Geschäftsführer: Maik Bachmann, Olaf Erb, Ralf Gebhart FreeBSD-Dienstleistungen, -Produkte und mehr: http://www.secnetix.de/bsd "If Java had true garbage collection, most programs would delete themselves upon execution." -- Robert Sewell ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Re: another error with md malloc based fs
Oliver, Thanks for this excellent article. I was trying to write about the same thing, but keep it simple enough to improve the chances of it crossing the language barrier weel enough to be useful. I believe you did a really good job of this and saved me several minutes of tying to do the same. I thought the warning in the man page was adequate (until 7.0 changes the default to swap backed), but, at least in Michael's case, I guess I was wrong. I guess the man page need to somehow make it clear that the memory used by malloc backed mds is a very limited resource. -- R. Kevin Oberman, Network Engineer Energy Sciences Network (ESnet) Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab) E-mail: [EMAIL PROTECTED] Phone: +1 510 486-8634 Key fingerprint:059B 2DDF 031C 9BA3 14A4 EADA 927D EBB3 987B 3751 pgpND9BHTEswl.pgp Description: PGP signature
Re: another error with md malloc based fs
Michael Schuh wrote: > Chuck Swiger wrote: > > Michael Schuh wrote: > > > i can't understand how malloc can eat all available > > > memory, i have 2Gigs of it ;-) > > > so it seems to me i know what i doing, if > > > i have 1,6 Gigs free Memory, and i say ok get me 750Megs from > > > my 1,6 Gigs of free Memory, what was faulty on that > > > > The two choices involve a swap-based RAMdisk, which can use all of > > the available physical RAM it needs to, since this memory is > > swappable, or a kernel-memory-based RAMdisk, which uses wired-down > > memory from within the kernel. > > Ok, so that is it harder to understand for me in the first time, > if i understand it now right the swap or file based memory backend > is also in the ram, but it get another way managed from kernels vm. > the malloc based ramdisk get's not reallly managed by the kvm, > but it underlays under the paging and swapping, and also > ba the "thread-killer" there shot's thread down it it get's to much > work for the system > > i hope i understand this right, it is important for me... Let me try to explain it with a little different words. The following is a bit simplified, but it should give you an idea about the advantages and disadvantages of each md disk type. First of all, _both_ "malloc" and "swap"-backed md disks are RAM disks. You don't have to worry that a swap-backed md disk will be created on your swap partiton. It's not. It will reside in RAM, just like a "malloc" md disk. However, the difference is that a malloc md disk uses memory from the KVM area (kernel virtual memory), which is hardwired into the RAM. It cannot be paged, and it is taken from the kernel's own memory pool, which is usually much smaller than your total available RAM. It's taken from the same pool of RAM that's used for network buffers, driver data and similar things. On the other hand, "swap-backed" md disks use regular memory, so to speak, just like a normal user process. That also means that they are pageable, which means that they can be paged to swap if necessary (if the systems runs low on RAM). That's what "swap-backed" means. Because of that, they don't have a size limit (well, they shouldn't be bigger than your RAM + swap, of course). If there is enough RAM, then they will stay completely in RAM and will _not_ touch your swap at all. For typical RAM disks, such as for /tmp, you should use a swap-backed md disk (it should be the default). The "malloc" type should be used only for rather small, special uses. I hope it's now a little clearer. Best regards Oliver -- Oliver Fromme, secnetix GmbH & Co. KG, Marktplatz 29, 85567 Grafing b. M. Handelsregister: Registergericht Muenchen, HRA 74606, Geschäftsfuehrung: secnetix Verwaltungsgesellsch. mbH, Handelsregister: Registergericht Mün- chen, HRB 125758, Geschäftsführer: Maik Bachmann, Olaf Erb, Ralf Gebhart FreeBSD-Dienstleistungen, -Produkte und mehr: http://www.secnetix.de/bsd "[...] one observation we can make here is that Python makes an excellent pseudocoding language, with the wonderful attribute that it can actually be executed." -- Bruce Eckel ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Re: another error with md malloc based fs
Michael Schuh schrieb: hi Chuck, hi @list me again, behind the scenes, i would take over my Server by an hoster from Linux to FreeBSD on the running system, while the hoster takes money for pressing 2 buttons and put a disk in my Server...so i create now a mfs based system that can bootet from disk and resides fully in ram. I have tryed out Colin's depenguinator, but it fails...some times now i would do the steps by hand thanks cheers michael Hi Michael! Some months ago I modified the depenguinator and made it work again. But sorry, I didn't yet get to write down all the steps involved (I am not a good writer). But I recently found another much more comfortable way using QEMU which I documented. Does your provider offer a linux rescue system? If yes, you can try this: http://www.vrdevelopers.de/content/view/45/40/1/5/lang,en/ I'll try to follow up with the modified depenguinator asap, but don't hold your breath... HTH, Manuel ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Re: another error with md malloc based fs
hi Chuck, hi @list me again, behind the scenes, i would take over my Server by an hoster from Linux to FreeBSD on the running system, while the hoster takes money for pressing 2 buttons and put a disk in my Server...so i create now a mfs based system that can bootet from disk and resides fully in ram. I have tryed out Colin's depenguinator, but it fails...some times now i would do the steps by hand thanks cheers michael 2007/3/23, Chuck Swiger <[EMAIL PROTECTED]>: On Mar 22, 2007, at 4:17 PM, Michael Schuh wrote: > i can't understand how malloc can eat all available > memory, i have 2Gigs of it ;-) > so it seems to me i know what i doing, if > i have 1,6 Gigs free Memory, and i say ok get me 750Megs from > my 1,6 Gigs of free Memory, what was faulty on that The two choices involve a swap-based RAMdisk, which can use all of the available physical RAM it needs to, since this memory is swappable, or a kernel-memory-based RAMdisk, which uses wired-down memory from within the kernel. You shouldn't or can't create huge kernel-based RAMdisks without expanding the amount of KVM available to the kernel...it's not reasonable to wire down hundreds of MB or RAM in almost all situations. > i hope it gave another way to use such ram-disk without using > swap or other harddisk based backendotoh my ram is not fully > useful > for me and i must, after twelve years of using freebsd, looking > back to > linux, and that i never would, only for using ramdisks If you wanted to create a 10MB RAMdisk, using kernel-based memory would be fine. For things which are much larger than that, the kernel's global VM management is using techniques like page-fault frequency heuristics and will do a better job of keeping active stuff cached in RAM (regardless of whether the stuff is part of a running process or part of your RAMdisk filesystem), if you let the kernel manage the situation than if you try to configure hundreds of MB which can only be used for a RAMdisk. Anyway, if you want to use Linux instead, nobody here is going to try to persuade you to do otherwise. Have fun, -- -Chuck -- === michael-schuh.net === Michael Schuh Preußenstr. 13 66111 Saarbrücken phone: 0681/8319664 mobil: 0177/9738644 @: [EMAIL PROTECTED] === Ust-ID: DE251072318 === ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Re: another error with md malloc based fs
Hi Chuck, 2007/3/23, Chuck Swiger <[EMAIL PROTECTED]>: On Mar 22, 2007, at 4:17 PM, Michael Schuh wrote: > i can't understand how malloc can eat all available > memory, i have 2Gigs of it ;-) > so it seems to me i know what i doing, if > i have 1,6 Gigs free Memory, and i say ok get me 750Megs from > my 1,6 Gigs of free Memory, what was faulty on that The two choices involve a swap-based RAMdisk, which can use all of the available physical RAM it needs to, since this memory is swappable, or a kernel-memory-based RAMdisk, which uses wired-down memory from within the kernel. Ok, so that is it harder to understand for me in the first time, if i understand it now right the swap or file based memory backend is also in the ram, but it get another way managed from kernels vm. the malloc based ramdisk get's not reallly managed by the kvm, but it underlays under the paging and swapping, and also ba the "thread-killer" there shot's thread down it it get's to much work for the system i hope i understand this right, it is important for me... You shouldn't or can't create huge kernel-based RAMdisks without expanding the amount of KVM available to the kernel...it's not reasonable to wire down hundreds of MB or RAM in almost all situations. > i hope it gave another way to use such ram-disk without using > swap or other harddisk based backendotoh my ram is not fully > useful > for me and i must, after twelve years of using freebsd, looking > back to > linux, and that i never would, only for using ramdisks If you wanted to create a 10MB RAMdisk, using kernel-based memory would be fine. For things which are much larger than that, the kernel's global VM management is using techniques like page-fault frequency heuristics and will do a better job of keeping active stuff cached in RAM (regardless of whether the stuff is part of a running process or part of your RAMdisk filesystem), if you let the kernel manage the situation than if you try to configure hundreds of MB which can only be used for a RAMdisk. Anyway, if you want to use Linux instead, nobody here is going to try to persuade you to do otherwise. Have fun, nop, i don't runaway, i don't disappear to run over to "dead rats" :-D i will stick to FreeBSD, than i believe it gave no better OS for me :-D i have many fun with my "lovely" BSD. at this time it gave only one situation in that i use linux and i hope this changes i the near future (VMWare:-( ) thanks for helping to see things a little bit more clear regards michael -- -Chuck -- === michael-schuh.net === Michael Schuh Preußenstr. 13 66111 Saarbrücken phone: 0681/8319664 mobil: 0177/9738644 @: [EMAIL PROTECTED] === Ust-ID: DE251072318 === ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Re: another error with md malloc based fs
On Mar 22, 2007, at 4:17 PM, Michael Schuh wrote: i can't understand how malloc can eat all available memory, i have 2Gigs of it ;-) so it seems to me i know what i doing, if i have 1,6 Gigs free Memory, and i say ok get me 750Megs from my 1,6 Gigs of free Memory, what was faulty on that The two choices involve a swap-based RAMdisk, which can use all of the available physical RAM it needs to, since this memory is swappable, or a kernel-memory-based RAMdisk, which uses wired-down memory from within the kernel. You shouldn't or can't create huge kernel-based RAMdisks without expanding the amount of KVM available to the kernel...it's not reasonable to wire down hundreds of MB or RAM in almost all situations. i hope it gave another way to use such ram-disk without using swap or other harddisk based backendotoh my ram is not fully useful for me and i must, after twelve years of using freebsd, looking back to linux, and that i never would, only for using ramdisks If you wanted to create a 10MB RAMdisk, using kernel-based memory would be fine. For things which are much larger than that, the kernel's global VM management is using techniques like page-fault frequency heuristics and will do a better job of keeping active stuff cached in RAM (regardless of whether the stuff is part of a running process or part of your RAMdisk filesystem), if you let the kernel manage the situation than if you try to configure hundreds of MB which can only be used for a RAMdisk. Anyway, if you want to use Linux instead, nobody here is going to try to persuade you to do otherwise. Have fun, -- -Chuck ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Re: another error with md malloc based fs
Hi, ok i don't know this joke, kbut i know the word's you have two options option a and option b. i can't understand how malloc can eat all available memory, i have 2Gigs of it ;-) so it seems to me i know what i doing, if i have 1,6 Gigs free Memory, and i say ok get me 750Megs from my 1,6 Gigs of free Memory, what was faulty on that that would imply for me i can't use my 2 Gigs of ram... and it also implies that you can put as many ram as you can find, and you can never use that ram, while a programm says:" hey guy you are not knowing what you doing.. YOU would use 750Megs of you 24Gigs Ram, no, you are dumb, you don't use this" i would use my 2 Gigs of ram, in that way i like :-D i would a very fast temporary filesystem in my ram. not more, not less. and the other question is i have a OS that say's i am a release, and the documentation say you can do this, who say don't do this? i have read the thread abaut malloc, but the errors described are not the samei have enough ram. (so show it to me :-D errare humanum est, is it?) i hope it gave another way to use such ram-disk without using swap or other harddisk based backendotoh my ram is not fully useful for me and i must, after twelve years of using freebsd, looking back to linux, and that i never would, only for using ramdisks thanks regards michael 2007/3/23, Kevin Oberman <[EMAIL PROTECTED]>: > Date: Thu, 22 Mar 2007 23:04:16 +0100 > From: "Michael Schuh" <[EMAIL PROTECTED]> > Sender: [EMAIL PROTECTED] > > Hi, > first i have 2GB of RAM, there are be 1,6G free > so it should enough toi make the following: > > # mdconfig -a -t mallo -s 750m -u 0 > # bsdlabel -w /dev/md0 > # bsdlabel -e /dev/md0 (create "a" slice to use all free respective first 16 > blocks) > # newfs -O2 -U /dev/md0a > mount /dev/md0a /mnt > # cd to_a_place/with/650MB_of_data/ > # tar -cf - |( cd /mnt/; tar -xf -) > after ~256MB transferred i get following errors, and the system crashes > horridble first it doesnt react of anything, then it does silently > reboot > without a shutdown...leaves all unclean > > extract from all.log > Mar 22 22:08:14 flying-dutchman kernel: > g_vfs_done():md0a[WRITE(offset=686899200 > , length=131072)]error = 28 > Mar 22 22:08:14 flying-dutchman kernel: > g_vfs_done():md0a[WRITE(offset=687030272 > , length=131072)]error = 28 > Mar 22 22:08:14 flying-dutchman kernel: > g_vfs_done():md0a[WRITE(offset=687161344 > , length=131072)]error = 28 > Mar 22 22:08:14 flying-dutchman kernel: > g_vfs_done():md0a[WRITE(offset=687292416 > , length=131072)]error = 28 > > > Any impressions, suggestions or limitations for md with malloc? > This errors occures also if i request to reserve the space by creation > > thanks > > regards > > michael > > -- > === michael-schuh.net === > Michael Schuh > Preußenstr. 13 > 66111 Saarbrücken > phone: 0681/8319664 > mobil: 0177/9738644 > @: [EMAIL PROTECTED] Old, old joke: Patient: Doctor! It hurts when I do this! Doctor: Don't do that. This is a well known issue with malloc backed md disks. Use swap backed ones. They will be the default in V7. They have all of the advantages of malloc backed mds without the possibility of eating all available memory and crashing the system. malloc backed md disks are very dangerous and should only be used in special circumstances and with great care. -- R. Kevin Oberman, Network Engineer Energy Sciences Network (ESnet) Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab) E-mail: [EMAIL PROTECTED] Phone: +1 510 486-8634 Key fingerprint:059B 2DDF 031C 9BA3 14A4 EADA 927D EBB3 987B 3751 -- === michael-schuh.net === Michael Schuh Preußenstr. 13 66111 Saarbrücken phone: 0681/8319664 mobil: 0177/9738644 @: [EMAIL PROTECTED] === Ust-ID: DE251072318 === ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Re: another error with md malloc based fs
> Date: Thu, 22 Mar 2007 23:04:16 +0100 > From: "Michael Schuh" <[EMAIL PROTECTED]> > Sender: [EMAIL PROTECTED] > > Hi, > first i have 2GB of RAM, there are be 1,6G free > so it should enough toi make the following: > > # mdconfig -a -t mallo -s 750m -u 0 > # bsdlabel -w /dev/md0 > # bsdlabel -e /dev/md0 (create "a" slice to use all free respective first 16 > blocks) > # newfs -O2 -U /dev/md0a > mount /dev/md0a /mnt > # cd to_a_place/with/650MB_of_data/ > # tar -cf - |( cd /mnt/; tar -xf -) > after ~256MB transferred i get following errors, and the system crashes > horridble first it doesnt react of anything, then it does silently > reboot > without a shutdown...leaves all unclean > > extract from all.log > Mar 22 22:08:14 flying-dutchman kernel: > g_vfs_done():md0a[WRITE(offset=686899200 > , length=131072)]error = 28 > Mar 22 22:08:14 flying-dutchman kernel: > g_vfs_done():md0a[WRITE(offset=687030272 > , length=131072)]error = 28 > Mar 22 22:08:14 flying-dutchman kernel: > g_vfs_done():md0a[WRITE(offset=687161344 > , length=131072)]error = 28 > Mar 22 22:08:14 flying-dutchman kernel: > g_vfs_done():md0a[WRITE(offset=687292416 > , length=131072)]error = 28 > > > Any impressions, suggestions or limitations for md with malloc? > This errors occures also if i request to reserve the space by creation > > thanks > > regards > > michael > > -- > === michael-schuh.net === > Michael Schuh > Preußenstr. 13 > 66111 Saarbrücken > phone: 0681/8319664 > mobil: 0177/9738644 > @: [EMAIL PROTECTED] Old, old joke: Patient: Doctor! It hurts when I do this! Doctor: Don't do that. This is a well known issue with malloc backed md disks. Use swap backed ones. They will be the default in V7. They have all of the advantages of malloc backed mds without the possibility of eating all available memory and crashing the system. malloc backed md disks are very dangerous and should only be used in special circumstances and with great care. -- R. Kevin Oberman, Network Engineer Energy Sciences Network (ESnet) Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab) E-mail: [EMAIL PROTECTED] Phone: +1 510 486-8634 Key fingerprint:059B 2DDF 031C 9BA3 14A4 EADA 927D EBB3 987B 3751 pgpAVKp241epy.pgp Description: PGP signature
