More Lazy writes with SSD environment

Question

Post reply

More Lazy writes with SSD environment

Latheesh NK

SSCertifiable

Points: 6559
More actions
December 27, 2010 at 3:18 am

#244668

When am doing a SSD evaluation against a Hard disk, I am experiencing a spike in Perfmon counter Lazy writer with SSD environment when compared with Hard disk. Most of the sites are commenting that the random writes would be better in Hard disk environment rather than SSD.

Could some one help me out to understand the same and the relation of Lazy writer impact on writing operations?

Viewing 6 posts - 1 through 5 (of 5 total)

You must be logged in to reply to this topic. Login to reply

Geoff A SSChampion Points: 11511 More actions · Answer 1

First you need to share what your entire test system looks like.

Does your SSD support TRIM?

Are you running 2008 R2?

Windows Server 2008 R2 support the TRIM function, which the OS uses when it detects that a file is being deleted from an SSD. When the OS deletes a file on an SSD, it updates the file system but also tells the SSD via the TRIM command which pages should be deleted. At the time of the delete, the SSD can read the block into memory, erase the block, and write back only pages with data in them. The delete is slower, but you get no performance degradation for writes because the pages are already empty.

Latheesh NK SSCertifiable Points: 6559 More actions · Answer 2

Thanks for your reply and herewith the details you asked and some more clarification.

My testing environment is Windows2008 R2 only.

However, the production is Windows2008 Entp64Bit. Would TRIM support on this machine?

I am not clear on the following statement "The delete is slower, but you get no performance degradation for writes because the pages are already empty. "

Could you also please guide me to understand, the Lazy writes occurs the reason because of the deletes from the memory as you suggested?

Oliiii SSCertifiable Points: 5328 More actions · Answer 3

Is the load on the DB you are comparing the same?

The increased amount of lazy write could just come from increased load (because the performance gets better, you write quicker, your counter increase)

You could also monitor the transactions/sec, sec/transaction and byte/sec in perfmon to get a more accurate view of what's going on.

Latheesh NK SSCertifiable Points: 6559 More actions · Answer 4

Latheesh NK

SSCertifiable

Points: 6559

December 28, 2010 at 3:54 am

#1266661

I do confirm the load (data) is similar for both test. The only one thing I need to confirm the TRIM functional support on the SSD. Am gona ask my provider for the same.

Nadrek SSC-Insane Points: 20039 More actions · Answer 5

+1 on monitoring throughput of the test set as well as counters.

If you can, run a sqlio test set against both your spindle drives and your SSD's; I like to use 8KB, 64KB, 256KB, and 1MB sizes against both random and sequential, and pick whatever outstanding IOs you like (I use 2 and 8).

FYI, I've seen what strongly appear to be controller/driver issues (both SSD's and spindles choked on a specific test condition), so I definitely recommend testing both.