SSD for Cache Drive

[JarDo]

Has anyone used an SSD for the Cache drive.  I'm thinking that since the purpose of the cache drive is to speed up writes to the array that a fast SSD might be a good idea.  What are your thoughts?

[bubbaQ]

A huge waste of resources.  Any decent conventional hard drive today will exceed the capacity of the LAN.  You will see no improvement in speed with an SSD over a good fast regular hard drive as cache.

 

Not to mention the fact you will quickly wear out the write cycles on an SSD using it as a cache drive.

[WeeboTech]

A number of us use the "cache drive as a possible warm spare" mentality.

I.E. Purchase a cache drive the size of your parity.

Then if any drive fails, you abandon cache and assign that drive to the failed slot so you can recover quicker.

[JarDo]

Yes...  Two very good points.  I completely understand and agree so far.  Does anyone else have input?

[boof]

Form factor is appealing physically. Given it's 2.5" and you could, if you were inclined, strap it anywhere in side your case without too much worry. Saves a previous 3.5" slot for a data disk.

 

As mentioned no real performance bonus in traditional use but if you are a person who likes using your cache drive as working space for additional programs (vmware guests and virtual disks, sabnzbd etc) you may see an improvement there.

[WeeboTech]

Form factor is appealing physically. Given it's 2.5" and you could, if you were inclined, strap it anywhere in side your case without too much worry. Saves a previous 3.5" slot for a data disk.

 

I use a PCI Slot Rafter in one of my machines.

 

http://www.newegg.com/Product/Product.aspx?Item=N82E16817984008&Tpk=slot%20rafter

[Seven]

A number of us use the "cache drive as a possible warm spare" mentality.

I.E. Purchase a cache drive the size of your parity.

Then if any drive fails, you abandon cache and assign that drive to the failed slot so you can recover quicker.

 

Could this "use the cache drive as a warm spare" idea be applied to any failed drive in the array (data drives too?) or does it only work for the parity?  I've been shying away from using a cache drive thus far but this "warm spare" idea may cause me to change my mind.

[Joe L.]

A number of us use the "cache drive as a possible warm spare" mentality.

I.E. Purchase a cache drive the size of your parity.

Then if any drive fails, you abandon cache and assign that drive to the failed slot so you can recover quicker.

 

Could this "use the cache drive as a warm spare" idea be applied to any failed drive in the array (data drives too?) or does it only work for the parity?  I've been shying away from using a cache drive thus far but this "warm spare" idea may cause me to change my mind.

It can be used for a replacement of ANY drive, not just the parity drive. (as long as it is equal to or greater then the size of the parity drive)

[Seven]

A huge waste of resources.  Any decent conventional hard drive today will exceed the capacity of the LAN.  You will see no improvement in speed with an SSD over a good fast regular hard drive as cache.

 

Seconded.  The bottleneck would be the LAN, not the SSD drive.  An expensive SSD drive would be way underutilized in this scenario.  Better to use a conventional hard drive for this instead.

[Seven]

A number of us use the "cache drive as a possible warm spare" mentality.

I.E. Purchase a cache drive the size of your parity.

Then if any drive fails, you abandon cache and assign that drive to the failed slot so you can recover quicker.

 

Could this "use the cache drive as a warm spare" idea be applied to any failed drive in the array (data drives too?) or does it only work for the parity?  I've been shying away from using a cache drive thus far but this "warm spare" idea may cause me to change my mind.

It can be used for a replacement of ANY drive, not just the parity drive. (as long as it is equal to or greater then the size of the parity drive)

 

Thanks Joe!  I just reallized that WeeboTech DID say in his post that it can be used to replace ANY drive in the array.  I haven't had my coffee yet today. 

[c234rmf]

I am using a SSD for cache.

I had a previous cache drive fail.  I  like the idea of having a drive that does not have to spin all the time for the cache.

It is not a barn burner but it works.

 

 

http://lime-technology.com/forum/index.php?topic=3976.msg35480#msg35480

[Stokkes]

The one big issue (aside from the lack of any real performance improvements) for using an SSD is the way the cache drive works.

 

You are consistently writing/deleting/writing/deleting/writing/deleting data off the cache drive on a nightly basis. If you are someone that uses your unRAID array quite regularly and frequently (on a daily basis) copy new data to it, you will most likely wear out your SSD drive a lot quicker than normal.

[WeeboTech]

I think if Tom compiles ext4 into the kernel, then trim support will be available.

Using ext4 for cache may help extend the life of an SSD.

 

I've read that if you write 20GB a day to an SSD, it will still last past 5 years.

On my Windows XP vmware instance, the XP disk is a container file on an SSD.

every day the uncorrectable sector count goes up.

Yet if I'm using 20GB out of 60GB, I bet I'll still have plenty of life on the SSD.

 

How many people write more then 20GB a day every single day?

[bubbaQ]

I've read that if you write 20GB a day to an SSD, it will still last past 5 years.

 

That seems widely optimistic based on what I've seen.  It depends on the SSD size, and whether it is MLC or SLC.

[WeeboTech]

I've read that if you write 20GB a day to an SSD, it will still last past 5 years.

 

That seems widely optimistic based on what I've seen.  It depends on the SSD size, and whether it is MLC or SLC.

 

I'm sure SSD size plays a big part in it.

From what I've read it's the ratio of size vs free space that has a big factor in it.

[bubbaQ]

From what I've read it's the ratio of size vs free space that has a big factor in it.

 

Partially... more specifically the problem is how many cells get locked up in static WORM files, as that reduces efficiency of wear leveling algorithms.