Simulating Real Random I/O with FIO

Simulating Real Random I/O with FIO

[Neto, Antonio Jose Rodrigues]

Hi All,

This is neto from Brazil

How are you?

I am doing a performance POC and I need to simulate 100% true random workload. I am using Fibre Channel and Windows 2008R2.

I have 8 LUNs (125GB each) and my configuration file is:

[workload]
bs=8k
ioengine=windowsaio
iodepth=3
numjobs=35
direct=1
runtime=120
size=1024g
filename=\\.\PhysicalDrive1
filename=\\.\PhysicalDrive2
filename=\\.\PhysicalDrive3
filename=\\.\PhysicalDrive4
filename=\\.\PhysicalDrive5
filename=\\.\PhysicalDrive6
filename=\\.\PhysicalDrive7
filename=\\.\PhysicalDrive8
zonesize=1024m
zoneskip=3g
rw=randrw
rwmixread=80
rwmixwrite=20
thread
unified_rw_reporting=1
group_reporting=1
randrepeat=0
norandommap

Questions:

1) I am assuming the working set is the size parameter that is 1TB. Is that correct?
2) What is the correlaction of the zonesize and zoneskip with the randomness? Adding those will help with the randomness of the workload or not? 
3) If I use zonesize=1024m and zoneskip=3g I have like 100K IOPS (coming from cache not disks) and if I use zonesize=10g and zoneskip=26g I have more realistic numbers. Could someone explain which is the a good example to use zonesize and zoneskip?
4) Is it possible with fio to specify per disk randomness area: For example: disk1 - start blocks 456 till blocks 3456, disk2 - start blocks 4567 to 103456

Thank you very much

All the best

neto

Re: Simulating Real Random I/O with FIO

[Jens Axboe]

On Mon, Feb 04 2013, Neto, Antonio Jose Rodrigues wrote:
> Hi All,
> 
> This is neto from Brazil
> 
> How are you?
> 
> I am doing a performance POC and I need to simulate 100% true random workload. I am using Fibre Channel and Windows 2008R2.
> 
> I have 8 LUNs (125GB each) and my configuration file is:
> 
> [workload]
> bs=8k
> ioengine=windowsaio
> iodepth=3
> numjobs=35
> direct=1
> runtime=120
> size=1024g
> filename=\\.\PhysicalDrive1
> filename=\\.\PhysicalDrive2
> filename=\\.\PhysicalDrive3
> filename=\\.\PhysicalDrive4
> filename=\\.\PhysicalDrive5
> filename=\\.\PhysicalDrive6
> filename=\\.\PhysicalDrive7
> filename=\\.\PhysicalDrive8
> zonesize=1024m
> zoneskip=3g
> rw=randrw
> rwmixread=80
> rwmixwrite=20
> thread
> unified_rw_reporting=1
> group_reporting=1
> randrepeat=0
> norandommap
> 
> Questions:
> 
> 1) I am assuming the working set is the size parameter that is 1TB. Is
> that correct?

Correct

> 2) What is the correlaction of the zonesize and zoneskip with the
> randomness? Adding those will help with the randomness of the workload
> or not? 

Your setting of zonesize=1g, skip=3g, will do random read/write within a
1gb zone (so from 0-1G initially), then move to the next zone (3G-4G),
etc. For a rotating drive, this will mean shorter seeks, since while it
is doing one zone, it will only seek within that zone.

> 3) If I use zonesize=1024m and zoneskip=3g I have like 100K IOPS
> (coming from cache not disks) and if I use zonesize=10g and
> zoneskip=26g I have more realistic numbers. Could someone explain
> which is the a good example to use zonesize and zoneskip?

Depends on what you want, really. With zonesize=10g, you are now seeking
in a 10x wider area than you were before.

> 4) Is it possible with fio to specify per disk randomness area: For
> example: disk1 - start blocks 456 till blocks 3456, disk2 - start
> blocks 4567 to 103456

You'd have to use different jobs. You above example uses 8 disks and
accesses them from all 35 jobs. You could split it a bit, so that you
have 8 sections, each accessing one disk. You could have 4 jobs in each
section, giving you a total of 32 jobs. Then you can set offset/size for
each job separately.

-- 
Jens Axboe

Re: Simulating Real Random I/O with FIO

[Neto, Antonio Jose Rodrigues]

On 2/6/13 8:10 AM, "Jens Axboe" <axboe@kernel.dk> wrote:

>On Mon, Feb 04 2013, Neto, Antonio Jose Rodrigues wrote:
>> Hi All,
>> 
>> This is neto from Brazil
>> 
>> How are you?
>> 
>> I am doing a performance POC and I need to simulate 100% true random
>>workload. I am using Fibre Channel and Windows 2008R2.
>> 
>> I have 8 LUNs (125GB each) and my configuration file is:
>> 
>> [workload]
>> bs=8k
>> ioengine=windowsaio
>> iodepth=3
>> numjobs=35
>> direct=1
>> runtime=120
>> size=1024g
>> filename=\\.\PhysicalDrive1
>> filename=\\.\PhysicalDrive2
>> filename=\\.\PhysicalDrive3
>> filename=\\.\PhysicalDrive4
>> filename=\\.\PhysicalDrive5
>> filename=\\.\PhysicalDrive6
>> filename=\\.\PhysicalDrive7
>> filename=\\.\PhysicalDrive8
>> zonesize=1024m
>> zoneskip=3g
>> rw=randrw
>> rwmixread=80
>> rwmixwrite=20
>> thread
>> unified_rw_reporting=1
>> group_reporting=1
>> randrepeat=0
>> norandommap
>> 
>> Questions:
>> 
>> 1) I am assuming the working set is the size parameter that is 1TB. Is
>> that correct?
>
>Correct
>
>> 2) What is the correlaction of the zonesize and zoneskip with the
>> randomness? Adding those will help with the randomness of the workload
>> or not? 
>
>Your setting of zonesize=1g, skip=3g, will do random read/write within a
>1gb zone (so from 0-1G initially), then move to the next zone (3G-4G),
>etc. For a rotating drive, this will mean shorter seeks, since while it
>is doing one zone, it will only seek within that zone.
>
>> 3) If I use zonesize=1024m and zoneskip=3g I have like 100K IOPS
>> (coming from cache not disks) and if I use zonesize=10g and
>> zoneskip=26g I have more realistic numbers. Could someone explain
>> which is the a good example to use zonesize and zoneskip?
>
>Depends on what you want, really. With zonesize=10g, you are now seeking
>in a 10x wider area than you were before.
>
>> 4) Is it possible with fio to specify per disk randomness area: For
>> example: disk1 - start blocks 456 till blocks 3456, disk2 - start
>> blocks 4567 to 103456
>
>You'd have to use different jobs. You above example uses 8 disks and
>accesses them from all 35 jobs. You could split it a bit, so that you
>have 8 sections, each accessing one disk. You could have 4 jobs in each
>section, giving you a total of 32 jobs. Then you can set offset/size for
>each job separately.
>
>-- 
>Jens Axboe
>
>--
>To unsubscribe from this list: send the line "unsubscribe fio" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html

Thank you so much my friend.

Could you please do me a big favor send me an example (worse for
mechanical disks (with zones, offsets, different areas etc)). I would like
to make a blog post about it and we also could add it in our examples in
FIO distribution. I just need the "perfect" random test for mechanical
disks. I have some doubts between the relation about zones and offsets. If
you could give me an example, I really appreciate.
>