Re: [Aoetools-dis­cuss­] Wr­ite-through cac­he

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 




Unfortunately, tests show

- write to vblade with directio ->drbd : unusable ( very-very slow). I don`t test this config for data loss.

- ggaoed with directIO -> drbd : data corruption while power off machine and switching target to another.




Чтв 07 Мар 2013 00:58:04 +0400, PV <pv2pv@xxxxxxxxx> написал:


Thank You, this is exactly what I need to understand.



http://stackoverflow.com/questions/5055859/how-are-the-o-sync-and-o-direct-flags-in-open2-different-alike


Gesendet: Mittwoch, 06. März 2013 um 20:51 Uhr
Von: PV <pv2pv@xxxxxxxxx>
An: aoetools-discuss@xxxxxxxxxxxxxxxxxxxxx
Betreff: Re: Wr­ite-through cache
Thank You and yangyuanzhi@xxxxxxxxx !
'man vblade' has answer .. but I had not understand what is
"O_DIRECT and O_SYNC mode for accessing  the  underlying block device"
Let me try ...

Срд 06 Мар 2013 12:14:20 +0400, Alexandre <alxgomz@xxxxxxxxx> написал:
I don't know much about blade, but have you tried other aoe targets? For instance I know ggaoed have an option to use directIO, thus disabling any caching (and decreasing performance I guess, although I never measured this precisely). This option typically ensure data constancy, which seems to be  what you're trying to achieve.

On 6 Mar 2013 02:22, "PV" <pv2pv@xxxxxxxxx> wrote:
Hello all,
I`ve tested this software and see, that I can`t turn off write cache. "Write cache off" can be a great feature for use with DRBD or so on.
Recently I`ve tested AOE with DRBD. Everything was working fine, but when I turn off primary server and then start AOE at secondary while writing data - there was data corruption.
iSCSI target has "write cache off" options and data was written correctly.
But as for me, AOE is faster, simplier and Windows client don`t hang while switching vblade from one server to another.
May be somebody can patch sources or tell me another way to turn off write cache?
Looking at the source of vblade-20, I see that this function perform write:
 
linux.c
...
int
putsec(int fd, uchar *place, vlong lba, int nsec)
{
        return pwrite(fd, place, nsec * 512, lba * 512);
}
...
So need to sync write queue before return?
May be using function
blk_queue_flush(q, REQ_FLUSH | REQ_FUA)
?
If it will be usefull, I can give additional information about versions, testing methods and so on.
 
PV.
 
 

 
------------------------------------------------------------------------------
Symantec Endpoint Protection 12 positioned as A LEADER in The Forrester
Wave(TM): Endpoint Security, Q1 2013 and "remains a good choice" in the
endpoint security space. For insight on selecting the right partner to
tackle endpoint security challenges, access the full report.
http://p.sf.net/sfu/symantec-dev2dev[http://p.sf.net/sfu/symantec-dev2dev]
_______________________________________________
Aoetools-discuss mailing list
Aoetools-discuss@xxxxxxxxxxxxxxxxxxxxx[Aoetools-discuss@xxxxxxxxxxxxxxxxxxxxx]
https://lists.sourceforge.net/lists/listinfo/aoetools-discuss




------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://p.sf.net/sfu/appdyn_d2d_mar
_______________________________________________
Aoetools-discuss mailing list
Aoetools-discuss@xxxxxxxxxxxxxxxxxxxxx
https://lists.sourceforge.net/lists/listinfo/aoetools-discuss

[Index of Archives]     [Linux ARM Kernel]     [Linux SCSI]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Security]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux