SSD Caching a HDD RAID array

Abdullah · March 2021

I am recently considering SSD caching a RAID-1 HDD array.

Looking up online, there are a few tools for SSD caching like lvmcache and bcache.

Does anyone have experience using these tools ? is it safe to cache a raid array ?
Being software solutions, will they work as expected?
These are some questions I had in mind.
Feel free to share any opinions/advice

thanks!

seriesn · March 2021

Do not cache storage unless it is a hardware based solution. Don't.

AaronSS · March 2021

@seriesn said:
Do not cache storage unless it is a hardware based solution. Don't.

+1,000,000 times this

rm_ · March 2021

If it's a RAID1, you should cache it with a RAID1 of SSDs as well, else a failure of the single SSD will cause data loss.
Unless you use it in the write-through mode, which will not deliver nearly as much of the performance benefit.

Abdullah · March 2021

@seriesn @rm_ thanks for the advice, that helps!

Abdullah · March 2021

I was thinking of using write-through/around mode.
RAID1 HDD + NVMe for caching

martijnk · March 2021

SSD caching is pointless unless you are doing a lot of synchronous writes (i.e. database writes). But most people use asynchronous data only anyway which is already cached in memory before it's written to disk. If you then cache it on another device, SSD, you will only increase your write latency and your performance will probably go down.

Also I wonder why somebody would ever SSD cache a RAID-1 array, since that consists of 2 disks. Why not make a RAID1 of two SSD disks then? Much more efficent.

Abdullah · March 2021

@martijnk said:
SSD caching is pointless unless you are doing a lot of synchronous writes (i.e. database writes). But most people use asynchronous data only anyway which is already cached in memory before it's written to disk. If you then cache it on another device, SSD, you will only increase your write latency and your performance will probably go down.

Also I wonder why somebody would ever SSD cache a RAID-1 array, since that consists of 2 disks. Why not make a RAID1 of two SSD disks then? Much more efficent.

Hi @martijnk
thanks for your thoughts!
I was considering read caching (better read speeds for frequently accessed files), using one SSD and a raid-1 array of two HDDs.

tomazu · March 2021

@martijnk said:
Also I wonder why somebody would ever SSD cache a RAID-1 array, since that consists of 2 disks. Why not make a RAID1 of two SSD disks then? Much more efficent.

in the old days SSDs were small and costly, while large spinning disks cheap and available :-)

martijnk · March 2021

@Abdullah said:

@martijnk said:
SSD caching is pointless unless you are doing a lot of synchronous writes (i.e. database writes). But most people use asynchronous data only anyway which is already cached in memory before it's written to disk. If you then cache it on another device, SSD, you will only increase your write latency and your performance will probably go down.

Also I wonder why somebody would ever SSD cache a RAID-1 array, since that consists of 2 disks. Why not make a RAID1 of two SSD disks then? Much more efficent.

Hi @martijnk
thanks for your thoughts!
I was considering read caching (better read speeds for frequently accessed files), using one SSD and a raid-1 array of two HDDs.

Well with RAID1 you already have read speed of 2 disks. Plus frequent accessed files will also be cached in mem, so more memory might be a better option.

It really all depends on your workload though. Let's say you are seeding 500 GB of torrents constantly and you cache those on a SSD it sure might help you out. I don't see a lot of other useful scenarios though. Because with SSD read caching it has to copy the data first from the slow hdd to the ssd which will probably take a lot of time as well.

But yeah it really all depends, you should probably test both setups. One with and one without SSD caching.

martijnk · March 2021

@tomazu said:

@martijnk said:
Also I wonder why somebody would ever SSD cache a RAID-1 array, since that consists of 2 disks. Why not make a RAID1 of two SSD disks then? Much more efficent.

in the old days SSDs were small and costly, while large spinning disks cheap and available :-)

Yeah figured that might be the reason

elliotc · March 2021

What is the different between SSD cached and 2TB SSD as swap?

ehab · March 2021

@tomazu said:
in the old days ...

in the old days before internet life was simple and i enjoyed playing outside.

Abdullah · March 2021

@martijnk said:

@Abdullah said:

@martijnk said:
SSD caching is pointless unless you are doing a lot of synchronous writes (i.e. database writes). But most people use asynchronous data only anyway which is already cached in memory before it's written to disk. If you then cache it on another device, SSD, you will only increase your write latency and your performance will probably go down.

Also I wonder why somebody would ever SSD cache a RAID-1 array, since that consists of 2 disks. Why not make a RAID1 of two SSD disks then? Much more efficent.

Hi @martijnk
thanks for your thoughts!
I was considering read caching (better read speeds for frequently accessed files), using one SSD and a raid-1 array of two HDDs.

Well with RAID1 you already have read speed of 2 disks. Plus frequent accessed files will also be cached in mem, so more memory might be a better option.

It really all depends on your workload though. Let's say you are seeding 500 GB of torrents constantly and you cache those on a SSD it sure might help you out. I don't see a lot of other useful scenarios though. Because with SSD read caching it has to copy the data first from the slow hdd to the ssd which will probably take a lot of time as well.

But yeah it really all depends, you should probably test both setups. One with and one without SSD caching.

Thanks! that might be the best way to know.

SSD Caching a HDD RAID array

Comments