Single SSD in production server without RAID ?

[nonsub]

Having a debate with a vendor who wishes to provide a production server equiped only with a single Intel 320 80GB SSD. Not going to get into details but it is a very proprietary purpose built box - we don't have a choice. Nightly full images to another drive will occur although this means 3-4hrs (including support getting onsite) to recover from a failure. Writes are expected to be around 15GB/day and the drive will likely never near filling. Vendor claims that the chances of failure of this single SSD are less than that of 2 X Raid 1 HDD. Personally I doubt there are any numbers to prove this. The other suggestion is that SSD in RAID have a higher likelyhood of failing at the same or near same time. This I can believe and is why if I was doing the setup with RAID 1 I would purposely hammer one drive with writes before installing or use a preworn drive.

For now We agreed to at least monitor wear and perhaps change the drive every 2 years even if within normal parameters.

Any feedback appretiated. Bottom line is as far as I am concerned nobody is putting out servers with single SSD drives and I want to know is if my concern is valid/warranted or not?

Thanks!

 

[LokutusofBorg]

RAID 1 is all about risk-management. If you can afford for this server to be down while you replace the drive, then you can accept the risk. If you cannot afford for this server to be down, then the risk is too much. It's as simple as that. Some dude's word that that SSD isn't going to fail is going to amount to what, exactly, if the SSD *does* in fact fail?

 

[philipma1957]

intel ssd has a .5 percent fail rate after a year. so a 1 in 200. I suggest having a ups as it will help protect your ssd. The intels had problems with power failure then rapid power up.

 

[nonsub]

RAID 1 is all about risk-management. If you can afford for this server to be down while you replace the drive, then you can accept the risk. If you cannot afford for this server to be down, then the risk is too much. It's as simple as that. Some dude's word that that SSD isn't going to fail is going to amount to what, exactly, if the SSD *does* in fact fail?

Thanks. Just say it is true and that 2 X HDD in RAID 1 are more likely to have both drives fail than a single good SSD drive or for sake of argument lets just say they are the same.. then in terms of risk management the risk is the same no?

Any idea how you calculate the odds of 2 hard drives (say they have a 4% AFR) failing at same time?

Edit.. that is a simple calculation. 4% AFR = 1/25. RAID 1 would be 1/25^2 = 1/625. This incorrectly assumes that the drive failures are statistically independant something that is probably wrong if the 2 drives are the same.

 

[beginner99]

Besides down-time, can you live with possible data loss? If you "only" have nightly updates you on average lose 12 h of data in case of failure.

 

[nonsub]

Besides down-time, can you live with possible data loss? If you "only" have nightly updates you on average lose 12 h of data in case of failure.

Again the question is is the risk any more than 2 mechanical drives in Raid 1?

 

[KentState]

Again the question is is the risk any more than 2 mechanical drives in Raid 1?

Both solutions to me are high risk. I would never implement a device in my organization for a critical function with either a single SSD or just 2 HDD in RAID 1. Now if the OS was only on the SSD/RAID 1 HDD then I may be a little more accepting, but the data and proprietary apps better be on a second channel in RAID 10 or RAID5 plus an online spare. If this is truly a server, then the cost of a second SSD for RAID 1 would not raise the cost substantially. Have you done an analysis of the risk of being without a server for a day or the loss of data? If we lost an email device for example, that could cost us $40k in revenue which is well worth the cost of redundancy.

 

[lsv]

A single drive for your production? No thanks. What if it fails in the middle of the day when it's needed most?

 

[blckgrffn]

Both solutions to me are high risk. I would never implement a device in my organization for a critical function with either a single SSD or just 2 HDD in RAID 1. Now if the OS was only on the SSD/RAID 1 HDD then I may be a little more accepting, but the data and proprietary apps better be on a second channel in RAID 10 or RAID5 plus an online spare. If this is truly a server, then the cost of a second SSD for RAID 1 would not raise the cost substantially. Have you done an analysis of the risk of being without a server for a day or the loss of data? If we lost an email device for example, that could cost us $40k in revenue which is well worth the cost of redundancy.

R5 won't protect you any better than two drives in R1.

The odds of two mechanical drives failing at the same time are pretty remote, as in very. Your biggest risk is going to be during rebuild time, so a hotspare is a must if you are being serious in your pursuit of uptime, availability and not getting angry calls from customers.

That said, any RAID controller worth having is going to be doing predictive failure analysis and is going to be pretty aggressive in using that HS. At that point, the "failed" drive is probably A-OK and data could be retrieved without issue.

Frankly, the ways that a SSD could crash and burn are pretty scary...

Anytime you need a server to stay up, it should use RAID protected volumes for its local data, end of story. This is true for your WHS and definitely anything called "production" even if you are just doing software RAID in the OS.

All of this is "IMHO" of course 🙂

 

[KentState]

R5 won't protect you any better than two drives in R1.

That's why I said "or". Depending on how many drives the server holds, RAID 10 or RAID 5 plus an online spare might be the max. I would just want a way to allow for 2 failures or a way to auto rebuild since it may not have phone home capabilities or someone to even replace a drive overnight or on weekends.