[ale] UFS/SDS v VxFS/VxVM (was: ext3 v reiserfs)
Dylan Northrup
docx at io.com
Mon Sep 15 15:01:54 EDT 2003
A long time ago, (15.09.03), in a galaxy far, far away, matty91 at bellsouth.n...:
:=Extent-based allocation really starts to shine when you start moving to
:=large databases (sequential and/or random access), big file servers, or
:=anything that can't be accessed directly with an inode (UFS still does
:=block based allocation, and following sindirect pointers takes time and
:=resources).
If I'm doing large databases I'd likely use raw I/O to a single LUN
presented by my SAN device that's got an appropriate RAID level and let it
go from there. I've not experienced systems where I/O wait times from using
a block-based FS was a noticeable performance hit, but I don't doubt there
are instances where this is the case.
:=VxFS also provides a more refined snapshot mechanism, online
:=extent reorganization, Quick I/O (provides pretty close to raw disk
:=performance), and a pretty slick clustered file system.
I agree with the points about snapshots, extent re-orgs (essentially
de-fragging the FS) and Quick I/O, but I'm wondering what clustered file
system you're talking about and what your definition is of a clustered file
system. You don't mean multiple machines accessing the same LUN, do you?
If so, I'd love a URL with a description.
:=If you have ever used Disksuite with disksets or large amounts of disk
:=devices, you would grow to love VxVM. Disksets are clunky, soft
:=partitioning is not real straight forward (VxVM DIDs rock),
With patch 108693-06 for SDS 4.2.1 and with Solaris 9's renaming of SDS to
SVM you get soft partitions. I haven't had to use them (when we use disks,
we generally use the entire disk or at least under 6 different file systems
per disk) but can see their facility.
:=and VxVM
:=is a piece of cake to manage (Disksutie is as well). I am not sure about
:=you, but how may times have you actually performed a Solaris upgrade?
I have performed patches that required I un-encapsulate my boot device, then
re-encapsulate it. I forget the details off-hand, but it has happened.
:=VxVM
:=costs money because it provides a ton of functionality and SAN management
:=capabilities not built into SDS.
Having not had to do proper SAN management (the joys of having a storage
group that won't let you touch their EMCs) I've only dealt with exposed
LUNs. When working with 20+ LLUNs, Disksuite would still be my choice. When
working with 50+ LUNs, maybe Veritas would be better. . . but I haven't had
to worry about that yet. The number of direct attached disks don't tend to
go that high and when using SAN disks I can carve out the LUNs to be
whatever size I like on the storage array.
The simplicity of SDS (now SVM) is what makes me like it better than VxVM.
The most hated aspect is the policy of Veritas to require node-locked
licensing and not allowing you to purchase a site license. For me, that's
reason enough right there. But what happens when your license expires and
you need to reboot your box? You're hosed . . . unless you can get a new
license from Veritas. These headaches are more trouble than I save by using
their software. But I'll admit that perhaps I haven't been asked to solve
the problem that Foundation is the proper solution for (and which UFS + SDS
isn't the proper solution for).
--
Dylan Northrup <*> docx at io.com <*> http://www.io.com/~docx/
"Harder to work, harder to strive, hard to be glad to be alive, but it's
really worth it if you give it a try." -- Cowboy Mouth, 'Easy'
More information about the Ale
mailing list