> 1) Mirroring protection or RAID 1 is NOT parity-based protection.
> RAID 1 provides 100% redundancy. Parity-based RAID does not, I
> repeat, DOES NOT provide 100% redundancy.
Did I say mirroring was parity-protected? Read this statement:
"This is important, because all customers that buy data
storage arrays, NAS or SANs, want parity-protection. Either
mirroring or some form of other RAID'd parity protection."
"Either mirroring or other RAID'd parity protection". Maybe
you could substitute "parity-protection" in the first sentence
for "protection". I don't want to confuse you.
> 2) Striping is RAID 0 so it's not, as you stubbornly insist,RAID
> off. RAID 0 is the best way to measure raw speed. Since most
> customers buy RAID boxes configured with multiple RAID levels
> involving trade-offs between protection and speed, EMC's approach
> is more reasonable. More importantly, your favorite benchmarks
> only measure one dimension: speed. You are avoiding the larger
> issue of why NTAP can't get its box certified for disaster-tolerance
> and fault-tolerance.
The "R" in RAID stands for redundancy. With RAID-0, striping,
there is no redundancy of data. RAID-0 really means "the absence
of RAID".
Yes, "RAID-0" is the best way to measure raw speed, because you
don't spend time having to copy data (mirroring) or parity-protect
with computation cycles like other RAID levels. NOTE: This means,
that EMC with RAID *OFF* (RAID-0), by your comment "the best way
to measure raw speed", is worse in SPEC SFS benchmarks than
NTAP's systems with RAID *ON*!
The benchmarks also measure availability Gus, if the submitting
company actually submits that way. EMC chooses to submit with
RAID *OFF*, no redundancy, so their numbers are not the performance
numbers for someone who wants a "highly available" system.
NTAP, Sun Microsystems, Hewlett Packard, IBM and even Auspex,
submit with RAID *ON*, to submit that their systems can
perform *WITH* availability features *enabled*.
So, if we can get a one sentence answer from you on this question:
1. Why does EMC submit SFS numbers with RAID *OFF*?
You would say: "because it is the best way to measure raw speed"?
If that's your answer, then that implies that the numbers EMC
submitted are "the best they can do" which further implies that
a customer who wants redundancy and performance is going to have
to spend *A LOT MORE MONEY* to get that level of performance and
redundancy with EMC products.
This, by the way, renders your argument of the 20% price difference
between the IP-4700 and F840 moot, since you are admitting you have
to buy a lot more components (disks, processors, memory) for the
IP-4700 to meet the performance of the NTAP 840. |
