Yet they don't list any of the manufacturers. Secondly, linode/digitalocean are supposed to be in a raid 10 (which usually includes a battery backup) which should limit any data loss (which both HDDs and SSDs would have). Not only that but both should have UPS with generator power. SSDs use lower power, have a lower MTBF rate and have better performance yet Linode customers will try to find some negative to justify why they are bad? This is just silly, outright silly. This shouldn't even be considered a study.

who said DO is supposed to be raid 10? they're not. You don't know they have BBUs either, you're just guessing.
The presentation which the article is based on listed 4 or 5 power outages suffered by major players in recent times, the UPS's failed.
I've heard SSDs fail more often than hard drives, but I haven't seen a study, feel free to cite one the shows SSDs last longer.
This is a linode forum for linode customers. You're surprised people here support linode?
The presentation and article aren't silly. The reason no manufacturers are listed are because they either want to sell that information or they can't release it due to manufactures contracts or some other reason.
DO have a slower network, provide less cores, have more VMs on a server, attract a different sort of customer that's more prone to abuse and attracts DDOS, have fewer support staff, have a smaller community and while this doesn't prove anything, they lost all of my data in less than a week. I had to rebuild from scratch.
I'm not a linode fanboi, I slam linode whenever I feel they deserve it.

I'm a Linode fanboi as well and you can ask the staff how much I support them and their efforts. What I don't support is trashing other technologies and companies to make us look good.

1) You are correct neither Linode nor DO states they use battery backup in their RAID configs yet both say they use raid for redundancy. I guess we will have to trust them both since we aren't there to physically verify.
2) Wait...so...because UPS supposedly failed; SSD failed?
Lets take a quote from the article "Modern storage technology (SSDs, No-SQL databases, commoditized RAID hardware, etc.) bring new reliability challenges to the already complicated storage stack. Among other things, the behavior of these new components during power faults—which happen relatively frequently in data centers"....Wait power faults happen regularly in Data Centers? Unless you are in Fremont I would guess you would reject this statement.
3) WAT? You heard from where? Start citing some sources because SSD MTBF beats HDD MTBF outright.
4) So you question point number one saying that we don't know for sure that DO has battery backup and then make an assertion here about contracts of the test without knowing the full story? Well that seems a bit silly.
5) You are full of crap because DOs network is faster than Linodes because they don't cap their customers. I can pull many examples if you would like to prove the point (still being a huge linode fanboi).
6) Sorry to hear that. Detail your experience so the rest can be warned with facts.

1) caker has said on irc they use raid 10 with BBU, and it also says on linode.com they use BBU and raid (not what type of raid). I've heard second hand DO use raid 5, I don't know about BBU, so I don't go around saying they have or they haven't got BBU. (unlike you)
2) I didn't say that, the study did though. you said DCs have UPSs, I pointed out that DCs still have power failures where the servers lose power suddenly with no chance to shut down gracefully. If you look at the presentation the article was based on, you will see a list of major players in major DCs suffering power outages recently. I can remember one, and it was amazon.
3) I heard from friends, and no thanks.
4) no you are
5) I can't believe what I'm reading, seriously, I actually lolled out loud. You see, the internet is like a series of tubes, it's not like a big truck you can just dump stuff on. The reasons you listed for DOs network being faster is the reason why it's slower, they have people on their network downloading TBs of data continuously for $5/month with NO cap, "unlimited", this slows everyone else down. Or you might be talking about linode's 50mbs out going rate? on my node in tokyo, I can burst to full 1G speed. Because linode charges a sustainable price for transfers, their network is much faster and consistently faster. They also have the money to increase capacity with demand. DO with their unlimited transfers will always be slower, their network, no matter how much capacity they add will always be saturated because people have "unlimited" transfers.
6) what details? I woke up to find my droplet was up and down and then just down, it was stuck trying to boot. I contacted support and they said they lost a raid set and my data was gone. You can't draw any conclusions from one data point though. a study from a university however...

I also would like to know what the 2 good manufacturers are.

RAID 10 does not include a battery backup, no level of RAID specifies a battery backup. Even the battery backed write cache you get on HP servers may be useless if the SSD lies about what it has put on stable storage. UPSs, generators, and the switching equipment between them fails. The sad fact is these things are not as reliable as they are claimed to be. I've seen diesel generators fail, UPSs die after a few seconds of runtime, switching equipment get stuck on or off. There is a reason EMC puts batteries in each and every symmetrix array.

SSDs fail in normal operation and when they do you don't even have the option of recovering anything, it's just all gone in one hit with no advance notice. DO better be using raid1 but I suspect they don't due to the extra costs involved.

How many times has a disk failed in a Linode host, Linode replaced it, and no customers even knew about it? I bet it happens all the time.

I do hope they are not cutting costs by using raid 5 on consumer grade SSDs.

Huh?

The paper was written for FAST '13 (the File and Storage Technologies Conference held in FEB 2013).

From what ass did you pull the 2003 figure?

The paper was done by postdoc's at OSU (Ohio State University) and by researchers at HP Labs.

They used 15 MODERN SSD drives from 5 MODERN vendors.

Geesh, the Infoworld article linked to all the technical papers - how hard was it to click on a few links and get your facts straight.

_________________
Either provide enough details for people to help, or sit back and listen to the crickets chirp.
Security thru obscurity is a myth - and really really annoying.

However, while we have over 50 years of collected wisdom working with spinning disk, flash-based SSDs are relatively new [1]

[1] Roberto Bez, Emilio Camerlenghi, Alberto Mod- elli, and Angelo Visconti. Introduction to Flash Memory. In Procedings of the IEEE, pages 489– 502, April 2003.

Go look at the sources the article uses....

I haven't used DigitialOcean, but I do have one question - when are they getting their own forums online?

@ohkus

If you don't understand how science (and engineering) papers are written, stick to the Unoobtu forums.

What part of "Introduction" is unclear - they site a reference so they don't have duplicate explaining to motards the basics of how flash memory works. The SSD tests were done in 2012. If you're going to spread FUD, at least do it right.

OMG - the paper uses DATA from 1990 - yes 1990 - yes 33 freaking years ago - so of course the study must be completely dated and bogus (the 1990 paper was by Gibson explaining the limitations of RAID, and it was his PhD thesis at UC Berkeley - Go Bears!).

_________________
Either provide enough details for people to help, or sit back and listen to the crickets chirp.
Security thru obscurity is a myth - and really really annoying.

1990 was not 33 years ago.

Sources: State of Iowa Department of Health, Record of Live Birth, 1980. Wegmans Bakery, Happy 0x20th Birthday Hoo Pycat, 2012.

_________________

/* TODO: need to add signature to posts */

Yes but 33 is a cooler number to rant about!

_________________
Either provide enough details for people to help, or sit back and listen to the crickets chirp.
Security thru obscurity is a myth - and really really annoying.

I skipped pages 4 through 11 of this thread. Did I miss anything?

_________________
Disclaimer: I am no longer employed by Linode; opinions are my own alone.

No.

_________________
Either provide enough details for people to help, or sit back and listen to the crickets chirp.
Security thru obscurity is a myth - and really really annoying.

Page 11 has a discussion of Yoda's bowel movements, surely that should be reviewed to ensure Linode's servers are running at optimim levels of performance.

James