iozone-ing an SSD (Re: Using an SSD "disk" for /)

[Ivan Voras]

On 11/06/10 20:00, Mikhail T. wrote:
> On 11/5/2010 7:13 PM, Mikhail T. wrote:
>> The results can be found in 4 HTML files found at:
>> http://aldan.algebra.com/~mi/io/ (The original iozone-created Excel
>> files are there too.)

That server doesn't respond!

> I added some more iozone runs, as well as those of rawio. These are much
> fewer (as file-system parameters don't affect rawio) and easier to
> interpret:
>
>      * It makes no difference to the SSD, whether your access is random
>        or sequential

And this is their biggest strength. All others - like "raw" IO speed, 
are in the majority of serious use cases secondary to that.

>      * SSD clearly beats the HD in rawrite, although, at "only" 88Mb/sec,
>        the results are far from the marketing...

Basically, you should be looking at IOPS, not MB/s.

>      * SSD connected to plain SATA port strongly beats the same SSD
>        connected to the fancy SAS controller (mpt)

Not really surprising. The controller might be too smart for its own 
good in this simple case.

But there could be other, more important resons, like the controller 
disabling the drive's (in this case, SSD's) write caching, which the 
majority of "real" RAID controllers do by default. Leaving the drive's 
write cache turned on puts your data at risk (which is important if you 
are running servers). You can sort of verify this hypothesis by setting 
this loader tunable:

hw.ata.wc=0

- this should disable the disk write cache for (S)ATA drives.