Is this something that should work? The assumption is that there is a dedicated raw SWAP slice and after install /tmp (which will be on /) will be unmounted and mounted to zpool/tmp (just like zpool/home) Thoughts on this? This message posted from opensolaris.org
Matt B wrote:> Is this something that should work? The assumption is that there is a dedicated raw SWAP slice and after install /tmp (which will be on /) will be unmounted and mounted to zpool/tmp (just like zpool/home) > > Thoughts on this?you are aware that /tmp by default resides in memory these days? putting /tmp on disk can have quite severe impact on performance. Michael -- Michael Schuster Recursion, n.: see ''Recursion''
Well, I am aware that /tmp can be mounted on swap as tmpfs and that this is really fast as most all writes go straight to memory, but this is of little to no value to the server in question. The server in question is running 2 enterprise third party applications. No compilers are installed...in fact its a super minimal Solaris 10 core install (06/06). The reasoning behind moving /tmp onto ZFS was to protect against the occasional misdirected administrator who accidently fills up tmp while transferring a file or what have you. As I said its a production server, so we are doing our best to insulate it from inadvertent errors When this server was build it was built with 8GB of swap on a dedicated slice. /tmp was left on / (root) and later mounted on a zpool. Is this dangerous given the server profile? Am i missing something here? Some other SUN engineers say that /tmp "is" swap and vice versa on Solaris, but my understanding is that my dedicated swap slice "is" swap and is not directly accessible. /tmp is just another filesystem that is happens to be mounted on a zpool with a quota so there is no fear of user/admin error. Based on how the system was setup is this a correct assertion? This message posted from opensolaris.org
On Fri, 23 Mar 2007, Matt B wrote:> The server in question is running 2 enterprise third party > applications. No compilers are installed...in fact its a super minimal > Solaris 10 core install (06/06). The reasoning behind moving /tmp onto > ZFS was to protect against the occasional misdirected administrator who > accidently fills up tmp while transferring a file or what have you. As > I said its a production server, so we are doing our best to insulate it > from inadvertent errorsIn that case, I think the easiest approach would be to use the "size" tmpfs mount option, which limits the amount of VM /tmp can use.> Is this dangerous given the server profile? Am i missing somethingDangerous? I think not. But most likely suboptimal. -- Rich Teer, SCSA, SCNA, SCSECA, OpenSolaris CAB member CEO, My Online Home Inventory Voice: +1 (250) 979-1638 URLs: http://www.rite-group.com/rich http://www.myonlinehomeinventory.com
On Fri, Mar 23, 2007 at 11:57:40AM -0700, Matt B wrote:> > The server in question is running 2 enterprise third party > applications. No compilers are installed...in fact its a super minimal > Solaris 10 core install (06/06). The reasoning behind moving /tmp onto > ZFS was to protect against the occasional misdirected administrator > who accidently fills up tmp while transferring a file or what have > you. As I said its a production server, so we are doing our best to > insulate it from inadvertent errorsYou can solve that problem by putting a size limit on /tmp. For example, we do this in /etc/vfstab: swap - /tmp tmpfs - yes size=500m The filesystem will still fill up, but you won''t run out of swap space. -- -Gary Mills- -Unix Support- -U of M Academic Computing and Networking-
Ok so you are suggesting that I simply mount /tmp as tmpfs on my existing 8GB swap slice and then put in the VM limit on /tmp? Will that limit only affect users writing data to /tmp or will it also affect the systems use of swap? This message posted from opensolaris.org
For reference...here is my disk layout currently (one disk of two, but both are
identical)
s4 is for the MetaDB
s5 is dedicated for ZFS
partition> print
Current partition table (original):
Total disk cylinders available: 8921 + 2 (reserved cylinders)
Part Tag Flag Cylinders Size Blocks
0 root wm 1 - 765 5.86GB (765/0/0) 12289725
1 swap wu 766 - 1785 7.81GB (1020/0/0) 16386300
2 backup wm 0 - 8920 68.34GB (8921/0/0) 143315865
3 var wm 1786 - 2550 5.86GB (765/0/0) 12289725
4 unassigned wm 2551 - 2557 54.91MB (7/0/0) 112455
5 unassigned wm 2558 - 8824 48.01GB (6267/0/0) 100679355
6 unassigned wm 0 0 (0/0/0) 0
7 unassigned wm 0 0 (0/0/0) 0
8 boot wu 0 - 0 7.84MB (1/0/0) 16065
9 unassigned wm 0 0 (0/0/0) 0
--------------
df output
--------------
df -k
Filesystem kbytes used avail capacity Mounted on
/dev/md/dsk/d0 6050982 1172802 4817671 20% /
/devices 0 0 0 0% /devices
ctfs 0 0 0 0% /system/contract
proc 0 0 0 0% /proc
mnttab 0 0 0 0% /etc/mnttab
swap 9149740 436 9149304 1% /etc/svc/volatile
objfs 0 0 0 0% /system/object
/usr/lib/libc/libc_hwcap2.so.1
6050982 1172802 4817671 20% /lib/libc.so.1
fd 0 0 0 0% /dev/fd
/dev/md/dsk/d3 6050982 43303 5947170 1% /var
swap 9149312 8 9149304 1% /var/run
zpool/home 4194304 91 4194212 1% /home
zpool/data 49545216 3799227 45745635 8% /data
zpool/tmp 49545216 55 45745635 1% /tmp
This message posted from opensolaris.org
On Fri, 23 Mar 2007, Matt B wrote:> Ok so you are suggesting that I simply mount /tmp as tmpfs on my > existing 8GB swap slice and then put in the VM limit on /tmp? Will thatYes.> limit only affect users writing data to /tmp or will it also affect the > systems use of swap?Well, they''d potentially be sharing the slice, so yes, that''s possible. If your (say) 1GB /tmp becomes full, only 7GB will remain for paging. However, if /tmp is empty, the whole 8GB will be available for paging. -- Rich Teer, SCSA, SCNA, SCSECA, OpenSolaris CAB member CEO, My Online Home Inventory Voice: +1 (250) 979-1638 URLs: http://www.rite-group.com/rich http://www.myonlinehomeinventory.com
Ok, since I already have an 8GB swap slice i''d like to use, what would be the best way of setting up /tmp on this existing SWAP slice as tmpfs and then apply the 1GB quota limit? I know how to get rid of the zpool/tmp filesystem in ZFS, but I''m not sure how to actually get to the above in a post-install scenario with existing raw swap Thanks This message posted from opensolaris.org
On Fri, 23 Mar 2007, Matt B wrote:> Ok, since I already have an 8GB swap slice i''d like to use, what > would be the best way of setting up /tmp on this existing SWAP slice as > tmpfs and then apply the 1GB quota limit?Have a line similar to the following in your /etc/vfstab: swap - /tmp tmpfs - yes size=1024m -- Rich Teer, SCSA, SCNA, SCSECA, OpenSolaris CAB member CEO, My Online Home Inventory Voice: +1 (250) 979-1638 URLs: http://www.rite-group.com/rich http://www.myonlinehomeinventory.com
And just doing this will automatically target my /tmp at my 8GB swap slice on s1 as well as placing the quota in place? This message posted from opensolaris.org
On Fri, 23 Mar 2007, Matt B wrote:> And just doing this will automatically target my /tmp at my 8GB swap > slice on s1 as well as placing the quota in place?After a reboot, yes. -- Rich Teer, SCSA, SCNA, SCSECA, OpenSolaris CAB member CEO, My Online Home Inventory Voice: +1 (250) 979-1638 URLs: http://www.rite-group.com/rich http://www.myonlinehomeinventory.com
Oh, one other thing...s1 (8GB swap) is part of an SVM mirror (on d1) This message posted from opensolaris.org
On Fri, 23 Mar 2007, Matt B wrote:> Oh, one other thing...s1 (8GB swap) is part of an SVM mirror (on d1)That''s not relevant in this case. -- Rich Teer, SCSA, SCNA, SCSECA, OpenSolaris CAB member CEO, My Online Home Inventory Voice: +1 (250) 979-1638 URLs: http://www.rite-group.com/rich http://www.myonlinehomeinventory.com