Re: Raid and SBS 2003
From: root (root_at_buchanangc.com)
Date: 02/17/04
- Next message: Dave Nelson: "Re: Will not boot without DVD"
- Previous message: Erik Likvarn: "POP3 Connector not delivering locally."
- In reply to: John LeMay: "Re: Raid and SBS 2003"
- Next in thread: root: "Re: Raid and SBS 2003"
- Reply: root: "Re: Raid and SBS 2003"
- Reply: John LeMay: "Re: Raid and SBS 2003"
- Messages sorted by: [ date ] [ thread ]
Date: Mon, 16 Feb 2004 19:02:19 -0800
"John LeMay" <jlemay@njmc.com> wrote in message
news:MPG.1a9b35e6ed8004ef98970c@news.individual.net...
> In article <OLPu57O9DHA.2416@TK2MSFTNGP10.phx.gbl>, root@buchanangc.com
> says...
> > Proper >hardware< RAID 5 will be nearly as fast as RAID 0 but require
one
> > more drive. www.3ware.com
> >
>
> You had me until this. What testing can you show that displays RAID5 can
> be nearly as fast as RAID0?
Don't need testing really as it's definitional. RAID 5 is striping +
parity. Writes can be slower but reads are about the same speed as RAID 0
because such a read IS a RAID 0 read but with one less drive.
> Note that cheating by caching writes doesn't
> count.
>
> I never understand why it's a surprise that RAID systems are slower than
> writing to a single disk.
Such is NOT slower than writing to single disk if a hardware RAID controller
is used and often not slower for software/firmware RAID 0 & 1.
> With RAID 1 systems - the fastest RAID that
> still provides redundancy
No, RAID 5 is as and usually faster than RAID 1 in most situations.
> - you are still writing the data twice.
Not true for hardware RAID. The host only sees one write. True however for
software/firmware but that wont be a significant issue in many cases.
> With
> RAID 5 you have to divide the data, stripe it across a lot of disks,
> then compute the parity and stripe it across the lot of disks as well
> (ok, the computation is probably done prior to either write, but the
> point remains). Writing data more than once, or writing extra data, is
> always going to be slower than writing data once.
No, because in RAID 0 and similarly RAID 5 those writes are done
concurrently. One has multiple heads on multiple drives putting data on the
disk surfaces simultaneously; that is the limiting process (where the heads
meets the oxide). Clearly RAID 5 parity writes do add overhead compared to
RAID 0.
A hardware RAID 5 4 drive array will write faster than a single drive.
> Even in RAID 0 systems
> there is still some overhead in deviding the data to be written,
A small amount of overhead for software/firmware RAID 0 and for hardware
RAID 0 there's no host CPU processing overhead.
> however
> these systems usually outperform a direct write since smaller amounts of
> data (and no extra data) is written to each disk in the array.
Huh?
>For this
> reason, several enterprises have selected RAID-10 (or 0+1) for heavily
> loaded datasets. The speed of RAID-0 plus RAID-1 (striping the data,
> then mirroring it) outperforms RAID-5 by a long shot and still allows
> for larger arrays than RAID-1.
That's false in most cases. Take eight drives and put it into a good
hardware RAID 5 array and in most cases that'll outperform a 4+4 RAID 0+1
array. AND the RAID 5 array is 1.75 times bigger. For cases where writing
vs reading is a high then the RAID 0+1 gains on the RAID 5 to the point
where RAID 0+1 is the preferred type for write only disk stores<g>.
A 16 drive RAID 0+1 array is faster than an 8 drive RAID 5 array. That may
be where the myth about the speed advantage of RAID 0+1 comes from. Take
those 16 drives and put them in a RAID 5 array and the balance swings back
to RAID 5.
- Next message: Dave Nelson: "Re: Will not boot without DVD"
- Previous message: Erik Likvarn: "POP3 Connector not delivering locally."
- In reply to: John LeMay: "Re: Raid and SBS 2003"
- Next in thread: root: "Re: Raid and SBS 2003"
- Reply: root: "Re: Raid and SBS 2003"
- Reply: John LeMay: "Re: Raid and SBS 2003"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
