Raid and different cache size

DrSteve

DrSteve

DrSteve

Associate
Joined
9 Nov 2004
Posts
592
Location
Northern Ireland
Hello.
I'm currently using an older 74Gb Raptor as a system drive - the 8MB cache version. I was toying with the idea of adding a second one and setting up raid 0. The new version of the drive has a 16MB cache and I was wondering if this makes any difference when setting up an array.
Thanks for any info.
 
Thanks for the replies. Have managed to source 2 more identical drives. Was initially going to go for raid 0 which I suppose I could still do but with more drives I suppose the chance of array failure is higher. Still trying to get my head round the various raid types. Am I right in thinking I could go for raid 5 with the three drives? Any point?
 
RAID 5 gives you both striping and distributed parity so you can lose a drive, but unless you have a controller with a dedicated processor it can be slow.
 
AbsenceJam said:
RAID 5 gives you both striping and distributed parity so you can lose a drive, but unless you have a controller with a dedicated processor it can be slow.
Click to expand...

Ahh ok. Was planning to use the built in controller on the mobo so I guess that means my own CPU will need to do all the work. Might just go for raid 0 and put the 3rd drive into another machine.
 
Good to know. :)
Just out of interest, how much of a strain would raid 5 be on the system? Given that it's an overclocked quad core would it be noticable or not?
 
DrSteve said:
Good to know. :)
Just out of interest, how much of a strain would raid 5 be on the system? Given that it's an overclocked quad core would it be noticable or not?
Click to expand...

probably not much, but it would be slow when writing to disc, which defeats the point of using the raptor's!

put them in raid 0 only for your OS and applications, then get a nice big drive for your data, that way if one dies you don't lose anything important.

but you should be doing backups anyway.
 
That's probably what I will do. Since I'm going to end up with 3 raptors I thought I would use 2 in raid 0 for windows and applications and keep the 3rd one for video editing. Recently got an HD camcorder so the extra speed of the raptor might come in handy there.
I'll probably add another 500GB drive to the quickly filling one I already have for data storage and backup.
 
RAID5 is terrible and IMO just an excuse for companies to sell overpriced and underperforming RAID Controllers to people who know no better.
In the days when a hard disk cost thousands it made some sense - not any more.
RAID0 is statistically *more* reliable than a single drive as it reduces the workload on the drives.
 
Got it all up and running this morning. Raid 0 with a 32K stripe. Vista certainly loads and shuts down faster. Getting an average transfer rate of 143Mb/sec from the array. Seems good so far. :)
 
FireShadow said:
RAID5 is terrible and IMO just an excuse for companies to sell overpriced and underperforming RAID Controllers to people who know no better.
In the days when a hard disk cost thousands it made some sense - not any more.
RAID0 is statistically *more* reliable than a single drive as it reduces the workload on the drives.
Click to expand...

You don't know what you're on about. RAID5 gives you striping with parity so you can lose a drive without losing any data, that has nothing to do with the price of a drive. The point of buying a decent RAID5 controller is that it speeds it up and you also get cache, hardware RAID5 with or without any cache is faster than RAID0.

Where are these statistics on RAID0 by the way? You may share the workload between two drives but the comparison to a single drive is odd, wouldn't it be better to compare to two single drives where RAID0 would work the drives far more?
 
think about how RAID5 works
every time you WRITE new data the parity has to be recalculated
this requires the data on the disk for that block to be READ
this means that writing a single 512 byte block to a RAID5 array
requires 64 blocks of data to be read - and then three blocks to be written across
at least three disks - during this process reading and writing to that stripe block
is locked
that is why RAID5 should never be used except for applications with very low write performance requirements
for example RAID5 is OK for a fileserver with a small group of people using office applications
but it simply will not work for any application which requires high performance
once the cache is saturated the performance of a RAID5 array drops to around 60th of a single hard drive
you can see this for yourself very easily - simply turn off the write cache on your RAID5 controller and benchmark it - it runs about as fast as a floppy drive
comparing RAID0 to spanning across two drives is pointless - spanning will thrash DISK0 and leave DISK1 unused until DISK0 starts to fill up - that is why spanning should never be used
there are three RAID levels that make sense IRL
- RAID0
- RAID1
- RAID10 or RAID0+1
once you realise that RAID5 has no value the next logical step is to realise that hardware RAID offers almost no real benefit over software RAID
the only feature hardware RAID can offer (block level caching) is available in software at much higher performance levels
either built in to applications (SQL Server) or as an add on product (SuperCache)
the reason these solutions outperform hardware RAID is simple - the caching is done in system RAM at 12GB/s and not over PCI/PCI-X/PCIe and no DMA
this means the amount of time taken to do data transfers is massively reduced - thus reducing contention and increasing overall performance of the server
which brings me back to the original point RAID5 is a rip-off that only still exists because so many people *think* it is good
and it gives raid card vendors the chance to make money by selling a totally pointless product
the worst mistake I see being made is people building database servers with RAID5 - absolute disaster
 
Last edited:
in reply to: mrn

"how do you store a 10TB database for example"
big database servers use RAID10 or RAID100
one array for data one for transaction logs - that is common knowledge

"what would you do if 2 drives failed?"
RAID10 with 4 drives can handle 2 drives failing as long as not from the same stripeset

At work I use a RAID6 storage server with 14 drives
it cost £20,000 and can not exceed 80MB/s throughput - about the same as a raptor!
That is RAID5/6 in action - a disaster!

"with raid5/6 you can have things like hot spares"
RAID10/100 also supports hot spares and live rebuilds

"anything implemented on hardware is going to be quicker than software"
I was referring to RAID levels which do *not* use parity
in which case the extra load on the CPU is so low as to be irrelevant

and parity calculations for a modern CPU at 100MB/s takes less than 2% of a single core anyway

RAID5 is history - just add one more drive and use RAID10 and you get:
- up to 50% redundancy (versus 33%)
- up to 2x write throughput (versus 0.5x)
- up to 4x read throughput (versus 2x)
- total flexibility - add extra drives and remove drives whenever you like
- not totally dependent on cache

I am not deluded - only the ignorant still use RAID5

here is quote from Microsoft Best Practices:

"In general, Raid-5 does *not* provide the best trade-off between reliability/availability and performance."

here is an example database server spec from some forum:

IBM x3650
Internal SAS RAID-1 2x73 GB (Operating System)

IBM DS3400 IBM EXP3000
QLogic 4Gb FC HBA (Primary) -----> Controller 1 -> SAS Controller
QLogic 4Gb FC HBA (Failover) -----> Controller 2 -> SAS Controller
RAID-10 8x 73 GB (Log Files)
RAID-1 2X 73 GB (Temp DB)
RAID-10 10X146 GB (Data Files)

notice all the RAID10 and complete lack of RAID5/6 ?
 
Last edited:
re: statistics on RAID0 versus single drive

the theory is simple
a hard disk has a probability of failure
that increases the more the drive is used

there are three types of activity that cause wear to a hard disk:
- spin up / down (very high wear)
- head movement (high wear)
- constant spinning (low wear)

as RAID0 reduces the writes to each drive by 50%
head movements are reduced by roughly 50% also

that means that probability of failure for each drive is reduced
in fact lifetime of each drive is almost doubled

if you draw a graph of probability of failure versus time/usage
of RAID0 versus single drive
it is slightly higher in the beginning
but probability of failure stays lower longer

thus RAID0 is actually slightly more reliable than a single drive
whilst giving up to 2x read and write throughput

the point is single drive configuration should be avoided

either use a stripe or mirror - never a single drive
 
going back to the OP question can you update us with the amount of drives you now have and what they are so we can advise of the best array for you to build.
from what Ive read of your post's you appear to have 1x 74gig raptor and have now sourced a further 2x36gig raptor
if this is so i would create a raid 01 also know as a mirrored stipped array
to do this raid the 2 36gig raptors in a stripe array (raid 0 ) and then use the 74gig raptor and mirror the striped array (raid 1)
this will give you the performance of a raid 0 array with the redundancy of a raid 1 array
that way if one of the 36 gig fail you will be able to replace the drive and repair the array using the 74 gig which has mirrored the data held on the other two.

the array's being discussed by the other poster's dont really pertain to the home pc user unless it is used to store mission critical data. and even then it would be more practical to use a secured VPN to access the data held at another location.

However back to your question if you give us the exact details of all the drive's you currently have we will be able to advise you propely.
 
You must log in or register to reply here.
Back
Top Bottom