SSD's in an Enterprise Server

[Stuka87]

So bit of background, we (Company I work for) have a FreeBSD 8.2 box that handles some network traffic and does not really require any amount of HD space.

A lot of people are wanting to put an SSD in it for "reliability". Which I can see from a vibration standpoint, as they will be shipped half way around the world. But there is zero need for the speed. I am suggesting a standard HD as they have a much more proven reliability history.

The model used would be a 40GB Intel 320.

My question is, how reliable is an SSD in a long term running situation, say 3-5 years down the road of running 24/7. Anybody know? Should there be any concern for failure. Be it mechanical, or drive corruption/controller issues?

Thanks!

 

[Coup27]

I dont think its a straight forward answer to be honest.

Moving parts vs no-moving parts is a no brainer.

SATA drives are only 'rated' for 8hrs continuous use per day. SSD's being non moving and all electronic I cannot see a reason why they can't run 24/7.

HDD firmware is a lot more mature than SSD firmware.

Intel adverised their failure rate ~0.4%. Can't see a HDD beating that.

All in all it all comes down to how stable the firmware and SSD are. Even what you would consider the most reliable SDD just had a major bug fixed a few weeks ago, and once those boxes are shipped it's out of your control.

You could also look at it from a business perspective. A small 5400rpm drive will be alot cheaper and a carries a 3yr warranty. What is the warranty of your product? If it's 3yrs or less, maybe theres your answer.

 

[Stuka87]

We cover them for 3 years. I know enterprise grade SATA drives are rated for 24/7 use, but have a smaller operational temperature window.

And that bug fix is the kind of thing that concerns me. HDD's rarely get firmware updates, as they don't change nearly as often. SSD's are still maturing.

 

[Phynaz]

Don't put a consumer drive in a server.

 

[Fallen Kell]

Don't put a consumer drive in a server.

This 1000%.

If you are going to put a SSD, make sure you use an enterprise SSD. If you use a HDD, use an enterprise HDD. This is a server, you don't want to have it fail simply because a disk died. That said as well, this is an ENTERPRISE SERVER! Use a fricking mirror like you should. NO enterprise class system should even not have a mirror running for their OS. Data, on-the-other-hand can be on a RAID 3, RAID 5, RAID 6, or RAID 0+1/1+0....

Hell, I have a 14 year old system still running 24x7 with the 2 ORIGINAL OS HDDs in it, and it is still humming along like it was 1997. I have replaced/upgraded the data drives, but the OS drives are still the original 18GB 10k RPM FastWide SCSI disks. And it still happily acts as my personal sFTP server (routible from the net), internal DNS, and LAMP stack. I have wiped the disks and re-formated many times over the years, but those disks keep on churning.

 

[razel]

A lot of people are wanting to put an SSD in it for "reliability". Which I can see from a vibration standpoint, as they will be shipped half way around the world.

I'm damm sure that when it's being shipped to the other side that it's not powered up and in use, so there's absolutely is no concern with any modern HDDs with shipment roughage --unless you are juggling it.

As for reliability, no one knows. However, Intel does have a history of playing safe even with chips that perform beyond their expected margins. It's a calculated risk (gamble) but that's business.

 

[MarkLuvsCS]

http://www.xtremesystems.org/forums/showthread.php?271063-SSD-Write-Endurance-25nm-Vs-34nm

Almost 500TB of data written to the nand. That is pretty impressive considering how long it has taken them to kill a single drive by hammering it for hours a day writing at the drives max speeds. Honestly seems SSDs will be very reliable for a 3-5 years without nand being a concern. This isn't even mentioning the fact it's a 64gb model so a 128gb should be able to last through twice as many writes. Controller is another complication though. Keep an eye on the thread and see how things pan out for the other drives.

 

[Stuka87]

http://www.xtremesystems.org/forums/showthread.php?271063-SSD-Write-Endurance-25nm-Vs-34nm

Almost 500TB of data written to the nand. That is pretty impressive considering how long it has taken them to kill a single drive by hammering it for hours a day writing at the drives max speeds. Honestly seems SSDs will be very reliable for a 3-5 years without nand being a concern. This isn't even mentioning the fact it's a 64gb model so a 128gb should be able to last through twice as many writes. Controller is another complication though. Keep an eye on the thread and see how things pan out for the other drives.

Great link, very interesting bit of data he has accumulated there. My main concern is the controller itself, but its good to see the NAND is holding up well.