How to properly benchmark dm-cache

How to properly benchmark dm-cache

[Remington Brasga]

[-- Attachment #1.1: Type: text/plain, Size: 525 bytes --]

Hi, I'm trying to benchmark dm-cache, but I'm having trouble getting the benchmark to show positive results. I've tried TPC-B, TPC-C, TPC-H, SQL, and a handful of other, but I think because they are simulations they don't generate the 'hot spots' required to promote blocks to the cache.

I know it's easy to show results using FIO with a specific configuration, but I was hoping to use a real benchmark like the ones I've tried.

Does anyone have a good set of benchmarks that shows positive results?

Thanks,
-Remi

[-- Attachment #1.2: Type: text/html, Size: 2644 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: How to properly benchmark dm-cache

[Mike Snitzer]

On Thu, Aug 31 2017 at  5:29pm -0400,
Remington Brasga <Remington.Brasga@taec.toshiba.com> wrote:

>    Hi, I'm trying to benchmark dm-cache, but I'm having trouble getting the
>    benchmark to show positive results. I've tried TPC-B, TPC-C, TPC-H, SQL,
>    and a handful of other, but I think because they are simulations they
>    don't generate the `hot spots' required to promote blocks to the cache.
> 
>    I know it's easy to show results using FIO with a specific configuration,
>    but I was hoping to use a real benchmark like the ones I've tried.
> 
>    Does anyone have a good set of benchmarks that shows positive results?

Hi Remi,

Which kernel are you testing?

I know Sanjay Rao (cc'd) has done various dm-cache testing with an OLTP
workload against an Oracle DB and that he has seen very solid results
with the latest dm-cache.

I'm not sure if he, or others, tried any of the TPC variants but
hopefully Sanjay can respond with where/how he got dm-cache to respond
well to his tests.

Mike

Re: How to properly benchmark dm-cache

[Remington Brasga]

Hi Mike,

I'm using the latest release of RHEL 7.4, kernel version 3.10.0-693.el7.x86_64.

Thanks for sharing. I am looking forward to Sanjay's comments.

Thanks,
-Remi

-----Original Message-----
From: Mike Snitzer [mailto:snitzer@redhat.com] 
Sent: Friday, September 01, 2017 08:59
To: Remington Brasga
Cc: dm-devel@redhat.com; Sanjay Rao; ejt@redhat.com
Subject: Re: How to properly benchmark dm-cache

On Thu, Aug 31 2017 at  5:29pm -0400,
Remington Brasga <Remington.Brasga@taec.toshiba.com> wrote:

>    Hi, I'm trying to benchmark dm-cache, but I'm having trouble getting the
>    benchmark to show positive results. I've tried TPC-B, TPC-C, TPC-H, SQL,
>    and a handful of other, but I think because they are simulations they
>    don't generate the `hot spots' required to promote blocks to the cache.
> 
>    I know it's easy to show results using FIO with a specific configuration,
>    but I was hoping to use a real benchmark like the ones I've tried.
> 
>    Does anyone have a good set of benchmarks that shows positive results?

Hi Remi,

Which kernel are you testing?

I know Sanjay Rao (cc'd) has done various dm-cache testing with an OLTP workload against an Oracle DB and that he has seen very solid results with the latest dm-cache.

I'm not sure if he, or others, tried any of the TPC variants but hopefully Sanjay can respond with where/how he got dm-cache to respond well to his tests.

Mike