Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
HIROSE, Masaaki wrote:
>Hi,
>
>in "Re: [DRBD-user] DRBD on ramdisk"
> <43021A1B.200 at arx.net>
>at Tue, 16 Aug 2005 19:53:47 +0300,
> tchatzi at arx.net wrote:
>
>
>
>>>> Is there a possibility of serious problem on DRBD on /dev/ramX?
>>>>
>>>>
>>>I don't know. We never really considered such usage.
>>>
>>>I know that someone one this list used to have his apache session data
>>>on a drbd on ram disk, and it seemd to work well for him.
>>>
>>>
>>>
>>>
>>That would be me :)
>>It does work quite reliably, but I suspect that it won't do what you
>>expect, if I read your post right.
>>Something like NFS mounted directories is probably what you need. But
>>you sure are welcome to drbd those...
>>
>>
>
>I want to NFS export DRBD directory. Tomcat is running on each NFS
>clients and Tomcat stores serialized session data into NFS mounted
>directory and share session data with another Tomcat.
>
>and I want to use memory base device (ramdisk, tmpfs) as backend of
>DRBD instead of hard disk.
>
>because of
> - more faster. Tomcat should process many accesses.
> - more troubleless. hard disk is breakable...
>
> //
>
>I have new trouble... I do stress test, then use% of /dev/drbd become
>100% so remove all files and directories but use% is still 100%...
>
> # df /mnt/drbd
> Filesystem 1K-blocks Used Available Use% Mounted on
> /dev/drbd1 258768 258756 12 100% /mnt/drbd
> # df -h /mnt/drbd
> Filesystem Size Used Avail Use% Mounted on
> /dev/drbd1 253M 253M 12K 100% /mnt/drbd
>
> # find /mnt/drbd -type f
> (no outpu)
> # fuser -mv /mnt/drbd/
> (no output)
>
>Is there potential problem in DRBD on ramdisk? I would be very
>grateful for any kind of information.
>
>
>way to setup /dev/drbd1
> boot with ramdisk_size=524288
> drbd.conf
> on drbd01 {
> device /dev/drbd1;
> disk /dev/ram3;
> address XXX.XXX.XXX.XXX:7789;
> meta-disk /dev/ram1[1];
> }
> mkfs.xfs -f -s size=1024 -b size=1024 -i size=512,maxpct=0 -d size=268435456 /dev/drbd1
>
>NFS server exports following configuration.
> /mnt/drbd XXX.XXX.XXX.XXX/255.255.255.0(rw,no_root_squash,async)
>
>NFS clients mounts following mount option.
> -o rw,soft,intr,rsize=8192,wsize=8192,noac
>
>stress test is following.
> - loop
> fork 3 processes. and each process do following in background
> repeat 1000 times
> cat 0 > 1
> rm -f 1
> - loop
> bonnie++ -s 32 -r 0 -n 20:4096:2048 -x 1
>
>I do stress test on three NFS clients coincidently.
>
>The 'used 100%' problem occurred about 9 hours later.
>
>
what does dumpe2fs say about the filesystem ?
the only problem I ever had with sessions on ramdisks (not drbd
specific) was running out of inodes.
I don't really know how tomcat stores sessions, I'm using mod_perl and
Apache::Session, but I suspect they're similar.
>There is no problem in case of NFS exports /dev/ram3 directry. (stress
>testing 24 hours)
>
>
>
>