ONTAP Discussions

Disable Caching on a Netapp Node

VIVEKNANDAVANAM
2,663 Views

I am doing some performance tests on a Netapp filer.

The cache hit percentage is 100% when I am hitting the volumes with high IOs.

Is there a way to disable caching on a node?

1 ACCEPTED SOLUTION

yannb
2,663 Views

If you're talking FlashCache, yes, you can disable it using

options flexscale.enable off

Now, is that really what you want to do? Cache hit is actually a good thing.

You might need to adapt your test to what you really want to know. From my experience, testing pure disk performances with no cache is a waste of time, you will not get any meanigful number that makes sense in a production context.

Even if you want to design a multi vendor test, you don't want to strip any features to achieve a common ground for all the arrays you're testing.

Implement best practices for all arrays, then run your test. How?

Well, the best test, as you certainly may have heard, is the test that uses your production data. You really should try to implement that, but I know most of the time you don't get the data, you can't align the right resource to perform the software installs, you don't have the capacity, etc etc...

So, in that case, try to qualify the workload the best you can : What is the data set (total amount of data for the workload), what is the working set (the actual amount of data used throughout the day), what is the read ratio, the nature of the IOs (sequential or random).

One you have these informations, or a rough estimate, you can use stress test tools to get numbers that makes sense. You can also use simple tools like "SIO" (available in the tool chest) that will let you run workload of a certain nature.

Also, trying to overload lots of SAS drives with a single server with a 1GB connection (for NAS) is not going to work. Make sure you have the gears to actually push the controllers.

As for your hit rate, it depends how you run the test... reading the same data that fits into your flash cache will not let you go to the disks. Use a tool like SIO to work on a test file that is at least 4 times four flash cache amount and to a lot of random IOs, to my knowledge, that would be the "worst case scenario" for flash cache. Oh, and consider reduplication... if you file is full of zeroes and you reduplicate, everything is going to fit in cache 🙂

My 2¢

View solution in original post

1 REPLY 1

yannb
2,664 Views

If you're talking FlashCache, yes, you can disable it using

options flexscale.enable off

Now, is that really what you want to do? Cache hit is actually a good thing.

You might need to adapt your test to what you really want to know. From my experience, testing pure disk performances with no cache is a waste of time, you will not get any meanigful number that makes sense in a production context.

Even if you want to design a multi vendor test, you don't want to strip any features to achieve a common ground for all the arrays you're testing.

Implement best practices for all arrays, then run your test. How?

Well, the best test, as you certainly may have heard, is the test that uses your production data. You really should try to implement that, but I know most of the time you don't get the data, you can't align the right resource to perform the software installs, you don't have the capacity, etc etc...

So, in that case, try to qualify the workload the best you can : What is the data set (total amount of data for the workload), what is the working set (the actual amount of data used throughout the day), what is the read ratio, the nature of the IOs (sequential or random).

One you have these informations, or a rough estimate, you can use stress test tools to get numbers that makes sense. You can also use simple tools like "SIO" (available in the tool chest) that will let you run workload of a certain nature.

Also, trying to overload lots of SAS drives with a single server with a 1GB connection (for NAS) is not going to work. Make sure you have the gears to actually push the controllers.

As for your hit rate, it depends how you run the test... reading the same data that fits into your flash cache will not let you go to the disks. Use a tool like SIO to work on a test file that is at least 4 times four flash cache amount and to a lot of random IOs, to my knowledge, that would be the "worst case scenario" for flash cache. Oh, and consider reduplication... if you file is full of zeroes and you reduplicate, everything is going to fit in cache 🙂

My 2¢

Public