RAID 5 write performance

[Tom Del Rosso]

Since it's striped I would expect RAID 5 to be faster than RAID 1 generally,
but I've read some vague references (in this link and elsewhere) to lower
write performance in 5 than in 1. Why would that be so?

http://www.msexchange.org/articles/Transaction-Logs-Lifeblood-Exchange.html

 

[Igor Batinic]

Hi!

Since it's striped I would expect RAID 5 to be faster than RAID 1 generally,
but I've read some vague references (in this link and elsewhere) to lower
write performance in 5 than in 1. Why would that be so?

http://www.msexchange.org/articles/Transaction-Logs-Lifeblood-Exchange.html

Indeed, RAID 5 (usually) is much slower that RAID 1 in such cases,
especially if you use RAID 5 array for such a purpose without a lot of
cache, battery backup etc.

RAID 5 is not recommended array for transaction logs not only in
Exchange environment, but also in all other environments.

Best regards,

Iggy

 

[Arno Wagner]

Since it's striped I would expect RAID 5 to be faster than RAID 1 generally,
but I've read some vague references (in this link and elsewhere) to lower
write performance in 5 than in 1. Why would that be so?

http://www.msexchange.org/articles/Transaction-Logs-Lifeblood-Exchange.html

Unkess you rite exactly in stripe size, in RAID5 you have to
read the rest of the stripe from the other drives, before
writing (to do the checksum). With RAID1 you just write,
regardless of size. This is mostly an issue for small accesses.

Arno

 

[Tom Del Rosso]

Hi!
http://www.msexchange.org/articles/Transaction-Logs-Lifeblood-Exchange.html

Indeed, RAID 5 (usually) is much slower that RAID 1 in such cases,
especially if you use RAID 5 array for such a purpose without a lot of
cache, battery backup etc.

RAID 5 is not recommended array for transaction logs not only in
Exchange environment, but also in all other environments.

But why is it slower in write mode? Does the controller write the parity
multiple times even when all the drives are modified?

 

[Tom Del Rosso]

Unkess you rite exactly in stripe size, in RAID5 you have to
read the rest of the stripe from the other drives, before
writing (to do the checksum). With RAID1 you just write,
regardless of size. This is mostly an issue for small accesses.

Thanks.

 

[Igor Batinic]

Hi!

But why is it slower in write mode? Does the controller write the parity
multiple times even when all the drives are modified?

Transaction (usually) are small records, and also usually each
transaction is written only to one disk in array. RAID 5 is most
efficient when "full stroke write" is usual.
That means: if small record is written to only one or two disks in
array, worst case scenario is to read data from all other disks,
calculate parity and then new data must be written to a couple disks in
array.
Therefore RAID 5 is not recommended array for any intensive I/O operation.

Best regards,

Iggy

 

[calypso]

Since it's striped I would expect RAID 5 to be faster than RAID 1 generally,
but I've read some vague references (in this link and elsewhere) to lower
write performance in 5 than in 1. Why would that be so?

http://www.msexchange.org/articles/Transaction-Logs-Lifeblood-Exchange.html

http://www.cs.berkeley.edu/~pattrsn/252F96/Lecture13.pdf

From one of the guys that invented RAID... Page 11 iz what you want to
see...


--
Mercedesa slusa u saboru plav jabukao udise danas. By runf

Damir Lukic, calypso@_MAKNIOVO_fly.srk.fer.hr
http://inovator.blog.hr
http://calypso-innovations.blogspot.com/

 

[Tom Del Rosso]

http://www.msexchange.org/articles/Transaction-Logs-Lifeblood-Exchange.html

http://www.cs.berkeley.edu/~pattrsn/252F96/Lecture13.pdf

From one of the guys that invented RAID... Page 11 iz what you want to
see...

Thanks a lot.