On Thu, March 24, 2011 12:30 pm, Volker Armin Hemmann wrote: > On Thursday 24 March 2011 08:49:52 J. Roeleveld wrote: >> On Wed, March 23, 2011 5:43 pm, Volker Armin Hemmann wrote: >> > md raid devices can do barriers. Don't know about lvm. But lvm is such >> a >> > can >> > of worms I am surprised people still recommend it. >> >> What is wrong with LVM? >> I've been using it successfully without any issues for years now. >> It does what it says on the box. > > it is another layer that can go wrong. Why take the risk? There > are enough cases of breakage after upgrades - and besides snapshots... is > the > amount of additional code running really worth it? Especially with bind > mounting?
There are always things that can go wrong and I agree, adding additional layers can increase the risk. However, the benefits of easily and quickly changing the size of partitions and creating snapshots for the use of backups are a big enough benefit to off-set the risk. Bind-mounting is ok, if you use a single filesystem for everything. I have partitions that are filled with thousands of small files and partitions filled with files are are, on average, at 1GB in size. I haven't found a filesystem yet that successfully handles both of these with identical performance. When I first tested performance I found that a simple "ls" in a partition would appear to just hang. This caused performance issues with my IMAP-server. I switched to a filesystem that better handles large amounts of small files and performance increased significantly. The way I do backups is that I stop the services, create snapshots and then restart the services. I then have plenty of time to backup the snapshot. If I were to do this differently, I'd end up having a downtime for over an hour just for a backup. Now, it's barely a minute of downtime. That, to me, is a very big bonus. -- Joost
