Re: > 4GB with NFS?
> On Thu, Jan 25, 2001 at 02:18:01PM -0800, Matthew Jacob wrote: > > An update on this > > > > If the server is Solaris, neither NetBSD nor FreeBSD (i386 or alpha) have a > > problem (as clients). > > > > The problem is therefore in some interaction between this server (see > > http://www.traakan.com- sorta like a NetApp) and *BSD. Hmm!!! > > > Did you check *BSD uses NFSv3 with this server ? Other than believing what it says on the server and and doing it manually, no. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
On Thu, Jan 25, 2001 at 02:18:01PM -0800, Matthew Jacob wrote: > An update on this > > If the server is Solaris, neither NetBSD nor FreeBSD (i386 or alpha) have a > problem (as clients). > > The problem is therefore in some interaction between this server (see > http://www.traakan.com- sorta like a NetApp) and *BSD. Hmm!!! Did you check *BSD uses NFSv3 with this server ? -- Manuel Bouyer <[EMAIL PROTECTED]> -- To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
> VLF-incapable utilities. (on a related note, is there a need for > vlfread()/vlfwrite() in the BSD's, or is VLF support native in the read/write > calls? The standard off_t is 64 bits in all of the BSDs. -- ... every activity meets with opposition, everyone who acts has his rivals and unfortunately opponents also. But not because people want to be opponents, rather because the tasks and relationships force people to take different points of view. [Dr. Fritz Todt] V I C T O R Y N O T V E N G E A N C E To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
> knowing NFS in general far better than *BSD in specific, I would guess the best > thing to do (if you suspect server/client communication anomaly) is to grab a > snoop/tcpdump of the failure. I'm trying to think of a clever way to cause the > failure immediately, so you're not tracing 4GB of writes... mebbe dd's > seek/skip? or just append to the existing 4GB file. > > also, what command are you using on the bsd's to write the 4GB file? I've lmdd or dd... > definitely seen issues with VLF-capable OS's failing to write past 2/4GB due to > VLF-incapable utilities. (on a related note, is there a need for > vlfread()/vlfwrite() in the BSD's, or is VLF support native in the read/write > calls? An update on this If the server is Solaris, neither NetBSD nor FreeBSD (i386 or alpha) have a problem (as clients). The problem is therefore in some interaction between this server (see http://www.traakan.com- sorta like a NetApp) and *BSD. Hmm!!! To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
* Nathan Parrish <[EMAIL PROTECTED]> [010125 13:19] wrote: > knowing NFS in general far better than *BSD in specific, I would guess the best > thing to do (if you suspect server/client communication anomaly) is to grab a > snoop/tcpdump of the failure. I'm trying to think of a clever way to cause the > failure immediately, so you're not tracing 4GB of writes... mebbe dd's > seek/skip? or just append to the existing 4GB file. > > also, what command are you using on the bsd's to write the 4GB file? I've > definitely seen issues with VLF-capable OS's failing to write past 2/4GB due to > VLF-incapable utilities. (on a related note, is there a need for > vlfread()/vlfwrite() in the BSD's, or is VLF support native in the read/write > calls? It has to do with the vm system casting values into 32bit variables, it's been a long time since I looked at this, if I can find it again I might be able to do something about it. To answer your question about VLF (which I had to guess at) assuming you mean Very Large Files: 1) yes some tools break on them, I don't have a list handy. 2) BSD has had native VLF support since 4.4-BSD. (off_t is 64bit) -- -Alfred Perlstein - [[EMAIL PROTECTED]|[EMAIL PROTECTED]] "I have the heart of a child; I keep it in a jar on my desk." To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
knowing NFS in general far better than *BSD in specific, I would guess the best thing to do (if you suspect server/client communication anomaly) is to grab a snoop/tcpdump of the failure. I'm trying to think of a clever way to cause the failure immediately, so you're not tracing 4GB of writes... mebbe dd's seek/skip? or just append to the existing 4GB file. also, what command are you using on the bsd's to write the 4GB file? I've definitely seen issues with VLF-capable OS's failing to write past 2/4GB due to VLF-incapable utilities. (on a related note, is there a need for vlfread()/vlfwrite() in the BSD's, or is VLF support native in the read/write calls? nathan --- Matthew Jacob <[EMAIL PROTECTED]> wrote: > > To be fair, I should say that the server is a 'specical' box. > > But an lmdd writing to a file in 250GB partition that I started from Solaris > last night is still going. The NetBSD && FreeBSD writes both stopped at 4GB. > I > suppose it still could be the server, but, well, it's hard to sell against > something that "just works"... .:-) > > > On Thu, 25 Jan 2001, Matthew Jacob wrote: > > > > > On Thu, 25 Jan 2001, Dan Nelson wrote: > > > > > In the last episode (Jan 25), Matthew Jacob said: > > > > I came across an embarrassing comparison last night- > > > > > > > > FreeBSD NFS clients (well, i386) stop writing files at 4GB. > > > > > > > > Solaris, with O_LARGEFILE options in the open arguments, does not. > > > > > > > > Does anyone here know what FreeBSD ought to be doing about this? Or > > > > have I missed something? There is no O_LARGEFILE in fcntl.h (it is > > > > present for Solaris, ConvexOS and some other platforms, I believe). I > > > > thought the *BSDs had > 32 bit file support? Or is it only for local > > > > filesystems? > > > > > > FreeBSD has 64-bit file offsets by default, which make -DLARGEFILE > > > hackery unnecessary. > > > > So I thought! > > > > > > > > Make sure you're using NFSv3 mounts (should be the default, but if not, > > > add "nfsv3" to the options column in fstab). I cross-mount FreeBSD, > > > Tru64, and Solaris boxes via NFS and can access large files on all > > > combinations of client and server. > > > > Huh. Interesting. The default showed up as a nfsv3 mount: > > > > 1/25 2:12 mountd/v3: granted 192.67.166.79 to /bob ro=0 uid0=60001 > > > > The solaris mount showed up as: > > > > 1/25 2:06 mountd/v3: granted 192.67.166.155 to /bob ro=0 uid0=60001 > > 1/25 2:06 nfs/tcp accepted 192.67.166.155,1023 > > > > I'll try an explicit v3 mount/tcp and see if it's better. > > > > > > > > > > > > To Unsubscribe: send mail to [EMAIL PROTECTED] > > with "unsubscribe freebsd-current" in the body of the message > > > __ Do You Yahoo!? Yahoo! Auctions - Buy the things you want at great prices. http://auctions.yahoo.com/ To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
> > Matthew Jacob writes: > > > > Same code compiled on Solaris is happy. > > Perhaps there's some braindamage in it. I'm afraid of something like: > > #ifdef Solaris > typedef filefoo u_int64_t; > #else > typedef filefoo u_int32_t; > #endif > I'll try with dd then,... let y'all know... To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
Matthew Jacob writes: > > Same code compiled on Solaris is happy. Perhaps there's some braindamage in it. I'm afraid of something like: #ifdef Solaris typedef filefoo u_int64_t; #else typedef filefoo u_int32_t; #endif ;) Drew To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
On Thu, 25 Jan 2001, Andrew Gallatin wrote: > > Matthew Jacob writes: > > > > I came across an embarrassing comparison last night- > > > > FreeBSD NFS clients (well, i386) stop writing files at 4GB. > > > > Solaris, with O_LARGEFILE options in the open arguments, does not. > > > > Does anyone here know what FreeBSD ought to be doing about this? > > Or have I missed something? There is no O_LARGEFILE in fcntl.h (it is present > > for Solaris, ConvexOS and some other platforms, I believe). I thought the > > *BSDs had > 32 bit file support? Or is it only for local filesystems? > > > > -matt > > Normal /bin/dd works fine between on 4.2-RELEASE i386s here. > > This is writing to a raid0 fs on a FreeBSD/i386 server using an nfsv3 > mount, udp, 8k read/write size: > > % dd if=/dev/zero of=bigfile bs=1024k count=5000 > 5000+0 records in > 5000+0 records out > 524288 bytes transferred in 471.673794 secs (5479 bytes/sec) > > This is writing to a software raid5 fs on a Solaris/x86 (2.8) server > using an nfsv3 mount, udp, 8k read/write size: > > % dd if=/dev/zero of=bigfile bs=1024k count=5000 > 5000+0 records in > 5000+0 records out > 524288 bytes transferred in 1165.859965 secs (4497007 bytes/sec) > > Maybe you should look at "lmdd" ? Maybe it is either buggy, or it was > compiled a long, long time ago? Same code compiled on Solaris is happy. > (btw, this is a virgin 4.2 install, with none of my nfs opts in it). Hmm. Well, everyone sez this oughta work so I'll go and figure out why it ain't for me... To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
Matthew Jacob writes: > > I came across an embarrassing comparison last night- > > FreeBSD NFS clients (well, i386) stop writing files at 4GB. > > Solaris, with O_LARGEFILE options in the open arguments, does not. > > Does anyone here know what FreeBSD ought to be doing about this? > Or have I missed something? There is no O_LARGEFILE in fcntl.h (it is present > for Solaris, ConvexOS and some other platforms, I believe). I thought the > *BSDs had > 32 bit file support? Or is it only for local filesystems? > > -matt Normal /bin/dd works fine between on 4.2-RELEASE i386s here. This is writing to a raid0 fs on a FreeBSD/i386 server using an nfsv3 mount, udp, 8k read/write size: % dd if=/dev/zero of=bigfile bs=1024k count=5000 5000+0 records in 5000+0 records out 524288 bytes transferred in 471.673794 secs (5479 bytes/sec) This is writing to a software raid5 fs on a Solaris/x86 (2.8) server using an nfsv3 mount, udp, 8k read/write size: % dd if=/dev/zero of=bigfile bs=1024k count=5000 5000+0 records in 5000+0 records out 524288 bytes transferred in 1165.859965 secs (4497007 bytes/sec) Maybe you should look at "lmdd" ? Maybe it is either buggy, or it was compiled a long, long time ago? (btw, this is a virgin 4.2 install, with none of my nfs opts in it). Drew To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
> > But I won't let it go! I was hoping to replace my Solaris box with either > > FreeBSD or NetBSD as my main home directory server. FreeBSD 4.2 panics part > > way through the first LADDIS runs I was using to test it with and I can't get > > NetBSD to start as a LADDIS client (I hadn't got the auth sorted out yet), so > > I've been trying to scrutinize a lot of things very closely in this area. > > This is an old problem that I think was never addressed. It has to > do with the vm system (on i386). It may be fixed now, so keep playing > a bit, if not open a problem report and someone ought to get to it. Yah, I was gonna do that once -current stabilized a bit. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
* Matthew Jacob <[EMAIL PROTECTED]> [010125 09:18] wrote: > > > On Thu, 25 Jan 2001, Dan Nelson wrote: > > > In the last episode (Jan 25), Matthew Jacob said: > > > On Thu, 25 Jan 2001, Dan Nelson wrote: > > > > Make sure you're using NFSv3 mounts (should be the default, but if not, > > > > add "nfsv3" to the options column in fstab). I cross-mount FreeBSD, > > > > Tru64, and Solaris boxes via NFS and can access large files on all > > > > combinations of client and server. > > > > > > Huh. Interesting. The default showed up as a nfsv3 mount: > > > > > > 1/25 2:12 mountd/v3: granted 192.67.166.79 to /bob ro=0 uid0=60001 > > > > > > The solaris mount showed up as: > > > > > > 1/25 2:06 mountd/v3: granted 192.67.166.155 to /bob ro=0 uid0=60001 > > > 1/25 2:06 nfs/tcp accepted 192.67.166.155,1023 > > > > > > I'll try an explicit v3 mount/tcp and see if it's better. > > > > I use UDP mounts as it's easier to debug traces. > > > > The next step is to start looking as packet dumps... When you say that > > the FreeBSD clients "stop writing", what error do they get? > > None. As in EOF. > > > Also, what > > version of FreeBSD are you using? > > 4.2 in this case. 1.5.1 for NetBSD. > > > Okay- I think from what you've said and what Thor said for NetBSD is that this > is "supposed to work!" (which accords with what *I* thought too). I'll do some > more debugging as I can (I'm in a project deadline for 3 projects this week). > > But I won't let it go! I was hoping to replace my Solaris box with either > FreeBSD or NetBSD as my main home directory server. FreeBSD 4.2 panics part > way through the first LADDIS runs I was using to test it with and I can't get > NetBSD to start as a LADDIS client (I hadn't got the auth sorted out yet), so > I've been trying to scrutinize a lot of things very closely in this area. This is an old problem that I think was never addressed. It has to do with the vm system (on i386). It may be fixed now, so keep playing a bit, if not open a problem report and someone ought to get to it. -- -Alfred Perlstein - [[EMAIL PROTECTED]|[EMAIL PROTECTED]] "I have the heart of a child; I keep it in a jar on my desk." To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
On Thu, 25 Jan 2001, Dan Nelson wrote: > In the last episode (Jan 25), Matthew Jacob said: > > On Thu, 25 Jan 2001, Dan Nelson wrote: > > > Make sure you're using NFSv3 mounts (should be the default, but if not, > > > add "nfsv3" to the options column in fstab). I cross-mount FreeBSD, > > > Tru64, and Solaris boxes via NFS and can access large files on all > > > combinations of client and server. > > > > Huh. Interesting. The default showed up as a nfsv3 mount: > > > > 1/25 2:12 mountd/v3: granted 192.67.166.79 to /bob ro=0 uid0=60001 > > > > The solaris mount showed up as: > > > > 1/25 2:06 mountd/v3: granted 192.67.166.155 to /bob ro=0 uid0=60001 > > 1/25 2:06 nfs/tcp accepted 192.67.166.155,1023 > > > > I'll try an explicit v3 mount/tcp and see if it's better. > > I use UDP mounts as it's easier to debug traces. > > The next step is to start looking as packet dumps... When you say that > the FreeBSD clients "stop writing", what error do they get? None. As in EOF. > Also, what > version of FreeBSD are you using? 4.2 in this case. 1.5.1 for NetBSD. Okay- I think from what you've said and what Thor said for NetBSD is that this is "supposed to work!" (which accords with what *I* thought too). I'll do some more debugging as I can (I'm in a project deadline for 3 projects this week). But I won't let it go! I was hoping to replace my Solaris box with either FreeBSD or NetBSD as my main home directory server. FreeBSD 4.2 panics part way through the first LADDIS runs I was using to test it with and I can't get NetBSD to start as a LADDIS client (I hadn't got the auth sorted out yet), so I've been trying to scrutinize a lot of things very closely in this area. -matt To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
To be fair, I should say that the server is a 'specical' box. But an lmdd writing to a file in 250GB partition that I started from Solaris last night is still going. The NetBSD && FreeBSD writes both stopped at 4GB. I suppose it still could be the server, but, well, it's hard to sell against something that "just works"... .:-) On Thu, 25 Jan 2001, Matthew Jacob wrote: > > On Thu, 25 Jan 2001, Dan Nelson wrote: > > > In the last episode (Jan 25), Matthew Jacob said: > > > I came across an embarrassing comparison last night- > > > > > > FreeBSD NFS clients (well, i386) stop writing files at 4GB. > > > > > > Solaris, with O_LARGEFILE options in the open arguments, does not. > > > > > > Does anyone here know what FreeBSD ought to be doing about this? Or > > > have I missed something? There is no O_LARGEFILE in fcntl.h (it is > > > present for Solaris, ConvexOS and some other platforms, I believe). I > > > thought the *BSDs had > 32 bit file support? Or is it only for local > > > filesystems? > > > > FreeBSD has 64-bit file offsets by default, which make -DLARGEFILE > > hackery unnecessary. > > So I thought! > > > > > Make sure you're using NFSv3 mounts (should be the default, but if not, > > add "nfsv3" to the options column in fstab). I cross-mount FreeBSD, > > Tru64, and Solaris boxes via NFS and can access large files on all > > combinations of client and server. > > Huh. Interesting. The default showed up as a nfsv3 mount: > > 1/25 2:12 mountd/v3: granted 192.67.166.79 to /bob ro=0 uid0=60001 > > The solaris mount showed up as: > > 1/25 2:06 mountd/v3: granted 192.67.166.155 to /bob ro=0 uid0=60001 > 1/25 2:06 nfs/tcp accepted 192.67.166.155,1023 > > I'll try an explicit v3 mount/tcp and see if it's better. > > > > > > To Unsubscribe: send mail to [EMAIL PROTECTED] > with "unsubscribe freebsd-current" in the body of the message > To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
In the last episode (Jan 25), Matthew Jacob said: > On Thu, 25 Jan 2001, Dan Nelson wrote: > > Make sure you're using NFSv3 mounts (should be the default, but if not, > > add "nfsv3" to the options column in fstab). I cross-mount FreeBSD, > > Tru64, and Solaris boxes via NFS and can access large files on all > > combinations of client and server. > > Huh. Interesting. The default showed up as a nfsv3 mount: > > 1/25 2:12 mountd/v3: granted 192.67.166.79 to /bob ro=0 uid0=60001 > > The solaris mount showed up as: > > 1/25 2:06 mountd/v3: granted 192.67.166.155 to /bob ro=0 uid0=60001 > 1/25 2:06 nfs/tcp accepted 192.67.166.155,1023 > > I'll try an explicit v3 mount/tcp and see if it's better. I use UDP mounts as it's easier to debug traces. The next step is to start looking as packet dumps... When you say that the FreeBSD clients "stop writing", what error do they get? Also, what version of FreeBSD are you using? -- Dan Nelson [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
On Thu, 25 Jan 2001, Dan Nelson wrote: > In the last episode (Jan 25), Matthew Jacob said: > > I came across an embarrassing comparison last night- > > > > FreeBSD NFS clients (well, i386) stop writing files at 4GB. > > > > Solaris, with O_LARGEFILE options in the open arguments, does not. > > > > Does anyone here know what FreeBSD ought to be doing about this? Or > > have I missed something? There is no O_LARGEFILE in fcntl.h (it is > > present for Solaris, ConvexOS and some other platforms, I believe). I > > thought the *BSDs had > 32 bit file support? Or is it only for local > > filesystems? > > FreeBSD has 64-bit file offsets by default, which make -DLARGEFILE > hackery unnecessary. So I thought! > > Make sure you're using NFSv3 mounts (should be the default, but if not, > add "nfsv3" to the options column in fstab). I cross-mount FreeBSD, > Tru64, and Solaris boxes via NFS and can access large files on all > combinations of client and server. Huh. Interesting. The default showed up as a nfsv3 mount: 1/25 2:12 mountd/v3: granted 192.67.166.79 to /bob ro=0 uid0=60001 The solaris mount showed up as: 1/25 2:06 mountd/v3: granted 192.67.166.155 to /bob ro=0 uid0=60001 1/25 2:06 nfs/tcp accepted 192.67.166.155,1023 I'll try an explicit v3 mount/tcp and see if it's better. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: > 4GB with NFS?
In the last episode (Jan 25), Matthew Jacob said: > I came across an embarrassing comparison last night- > > FreeBSD NFS clients (well, i386) stop writing files at 4GB. > > Solaris, with O_LARGEFILE options in the open arguments, does not. > > Does anyone here know what FreeBSD ought to be doing about this? Or > have I missed something? There is no O_LARGEFILE in fcntl.h (it is > present for Solaris, ConvexOS and some other platforms, I believe). I > thought the *BSDs had > 32 bit file support? Or is it only for local > filesystems? FreeBSD has 64-bit file offsets by default, which make -DLARGEFILE hackery unnecessary. Make sure you're using NFSv3 mounts (should be the default, but if not, add "nfsv3" to the options column in fstab). I cross-mount FreeBSD, Tru64, and Solaris boxes via NFS and can access large files on all combinations of client and server. -- Dan Nelson [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
> 4GB with NFS?
I came across an embarrassing comparison last night- FreeBSD NFS clients (well, i386) stop writing files at 4GB. Solaris, with O_LARGEFILE options in the open arguments, does not. Does anyone here know what FreeBSD ought to be doing about this? Or have I missed something? There is no O_LARGEFILE in fcntl.h (it is present for Solaris, ConvexOS and some other platforms, I believe). I thought the *BSDs had > 32 bit file support? Or is it only for local filesystems? -matt To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
