Re: Patches for updated gvinum

On man, jul 09, 2007 at 12:00:52 +0200, Fluffles wrote:
> Ulf Lilleengen wrote:
> >Hi,
> >
> >Gvinum just got better. At least, that's what I hope. Lukas rewrote gvinum 
> >to
> >use a simpler and easier to handle single-thread event system, and I have 
> >been
> >adapting existing code to this. The result can be found in the patches 
> >here:
> >  
> 
> Nice to see gvinum get's the facelift it so desperately needed. It's 
> unfortunate gvinum was never able to recover from the GEOM-age, until 
> now maybe. :-)
> 
> However, do you think that it's performance is affected too? Last time i 
> tested, i was not impressed with gvinum's RAID5 performance, write 
> speeds were very low. RAID0 and RAID1 appeared to be on par with gstripe 
> and gmirror though. I guess gvinum is primary for people who feel 
> comfortable with its configuration and user-interface.
> 
It might, since it won't have pass BIOs through the consumer/provider system. I
did benchmark it a bit with gmirror with rawio, and gvinum was a bit faster. I
did a test on another system that showed gmirror a bit faster, so I guess
they're still the same. I'll try make a comparison with old gvinum as well. 

I heard about the RAID-5 problems, but I do not think that will be any better,
as the problem was related to the request ordering or at least the way RAID-5 is
done(I can't remember exactly, since it's in an old mailbox). However, this
seems to be a good tip on what I can look into next after my stability testing.

As you say, gvinum is primary for the people who like the complete "volume-manager" 
it offers. Also, what is important, is that people can upgrade old boxes running
vinum.

> >Therefore, I hope many will
> >test this patch and help discovering bugs that otherwise would just stay 
> >there.
> >  
> 
> I will be able to test this on my new test system, coming Wedsnesday. 
> This consists of a modern dualcore AMD 3GHz processor with 4GB RAM and 8 
> SATA disks. If you like, i can report my findings to the list when i'm done.
> 
Yes, that would be nice. Thank you very much, I appreciate your help! Currently,
I only have 6 "old" SCSI disks to run my real-world tests on.

-- 
Ulf Lilleengen