zfs discuss - May 2007 - ZFS vs UFS performance measurement

Hi folks,

 We have the following disks :and we want to create a STRIPE

 c7t2d0 c7t3d0 c7t4d0 c7t5d0 c7t8d0 c7t9d0 c8t2d0  c8t3d0 c8t4d0 c8t5d0 c8t8d0
c8t9d0

What we would like to measure is how the following two STRIPES perform
STRIPE ( Created using Solaris Volume Manager ) 
STRIPE created with ZFS. 

How can I acheive the exact STRIPE ( w.r.t to the interleaving, stripe size,
etc... ). We want to make sure that the two STRIPE configurations are identical.
Any pointers ?

Thanks in Advance
_D
 
 
This message posted from opensolaris.org

Durga Deep Tirunagari wrote:
> Hi folks,
> 
>  We have the following disks :and we want to create a STRIPE
> 
>  c7t2d0 c7t3d0 c7t4d0 c7t5d0 c7t8d0 c7t9d0 c8t2d0  c8t3d0 c8t4d0 c8t5d0
c8t8d0 c8t9d0
> 
> What we would like to measure is how the following two STRIPES perform
> STRIPE ( Created using Solaris Volume Manager ) 
> STRIPE created with ZFS. 
> 
> How can I acheive the exact STRIPE ( w.r.t to the interleaving, stripe
size, etc... ). We want to make sure that the two STRIPE configurations are
identical. Any pointers ?
This is not possible because SVM uses a fixed stripe allocation and ZFS
uses dynamic stripe allocation.  Reads and writes for ZFS will only use
the number of disks needed, not all of the disks (every time) like SVM.

In other words, it is like comparing apples and oranges, at least as far
as RAID implementations is concerned.
  -- richard

Durga.Tirunagari at Sun.COM wrote:
> Richard Elling wrote:
> 
>> Durga Deep Tirunagari wrote:
>>
>>> Hi folks,
>>>
>>>  We have the following disks :and we want to create a STRIPE
>>>
>>>  c7t2d0 c7t3d0 c7t4d0 c7t5d0 c7t8d0 c7t9d0 c8t2d0  c8t3d0 c8t4d0 
>>> c8t5d0 c8t8d0 c8t9d0
>>>
>>> What we would like to measure is how the following two STRIPES
perform
>>> STRIPE ( Created using Solaris Volume Manager ) STRIPE created with
ZFS.
>>> How can I acheive the exact STRIPE ( w.r.t to the interleaving, 
>>> stripe size, etc... ). We want to make sure that the two STRIPE 
>>> configurations are identical. Any pointers ?
>>
>>
>> This is not possible because SVM uses a fixed stripe allocation and ZFS
>> uses dynamic stripe allocation.  Reads and writes for ZFS will only use
>> the number of disks needed, not all of the disks (every time) like SVM.
>>
>> In other words, it is like comparing apples and oranges, at least as
far
>> as RAID implementations is concerned.
>>   -- richard
> 
> 
> hi Richard,
> 
> Please suggest an alternative ?. Please advice on whats the best course 
> of action
What are you trying to accomplish?
The PAE group in Sun has a team working on ZFS performance and characterization.
Some results have been blogged externally.  It may be that your question is
already answered someplace.
  -- richard

Richard Elling wrote:
> Durga Deep Tirunagari wrote:
>
>> Hi folks,
>>
>>  We have the following disks :and we want to create a STRIPE
>>
>>  c7t2d0 c7t3d0 c7t4d0 c7t5d0 c7t8d0 c7t9d0 c8t2d0  c8t3d0 c8t4d0 
>> c8t5d0 c8t8d0 c8t9d0
>>
>> What we would like to measure is how the following two STRIPES perform
>> STRIPE ( Created using Solaris Volume Manager ) STRIPE created with
ZFS.
>> How can I acheive the exact STRIPE ( w.r.t to the interleaving, 
>> stripe size, etc... ). We want to make sure that the two STRIPE 
>> configurations are identical. Any pointers ?
>
>
> This is not possible because SVM uses a fixed stripe allocation and ZFS
> uses dynamic stripe allocation.  Reads and writes for ZFS will only use
> the number of disks needed, not all of the disks (every time) like SVM.
>
> In other words, it is like comparing apples and oranges, at least as far
> as RAID implementations is concerned.
>   -- richard

hi Richard,

Please suggest an alternative ?. Please advice on whats the best course 
of action

_D