100 Samsung 840's

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

JSchuricht

Active Member
Apr 4, 2011
198
74
28
I just got a new project at work and need some help with convincing management on how bad of an idea this is. Basically someone gave me about 100 Samsung 840 drives and wants some older servers upgraded. The servers are running Supermicro SMC AOC-USAS-S8iR controllers on a passthru backplane, alternative is an Intel ICH9r onboard SATA connection. The drives would be in RAID1, 2 per server.

The issue I foresee is these are manufacturing servers running windows XP performing extremely time critical tasks. There are lots of small writes but only around 10TB/yr. My concern is lack of trim will lead to a stall during a write and cause a server to miss it's target time during an operation.

If anyone has some good hard data I can use to show this being an issue, I would appreciate it.
 

dba

Moderator
Feb 20, 2012
1,477
184
63
San Francisco Bay Area, California, USA
I found that the Samsung 830 and 840 Pro drives (not sure if your 840s are Pro or not) have pretty good garbage collection. I use them in large RAID setups without TRIM with no problems. The key is to leave tons of empty space. If I could format the drives to half of their capacity, and if the usual IOPS load was moderate (say <1000/s) then I'd be willing to use those drives. With those 2007-era IO chip, the servers can't be generating that much IO anyway!
 

JSchuricht

Active Member
Apr 4, 2011
198
74
28
The box hasn't made it to my desk yet so I am still trying to find the exact model number of the drives so I don't know if they are pro, evo, etc yet. You are correct, IO load is low, the servers currently have just two 300gb 15k drives mirrored.

If they do turn out to be pro drives, do you think the garbage collection is good enough to trust when a delayed write could add up to millions of dollars? Perhaps I am just being paranoid from 2008 era Jmicron controllers bringing a system to a stop when they ran out of empty cells.
 

dba

Moderator
Feb 20, 2012
1,477
184
63
San Francisco Bay Area, California, USA
If your app works fine on 150 IOPS spinning drives then an over-provisioned SSD drive with decent garbage collection will be just fine. I would expect max latency to be less than 2ms in the worst of scenarios - compared to WAY more than that if you let an older SSD drive fill up completely without TRIM. That said, if the servers are that consequential, then an Enterprise-grade drive with power fail protection - and perhaps even some newer servers as well - could be justified.
 
  • Like
Reactions: Patrick

JSchuricht

Active Member
Apr 4, 2011
198
74
28
Thanks, hopefully the drives turn out to be large enough to leave some free space. New servers are out for many reasons, the manufacturer of the tools is still shipping with the same old server so we are stuck in that regard. Power fail protection is a hard sell to management, the entire site has redundant power which is about as far as that conversation will go. On the plus side, the 840's were all free, just need to convince the right person to grab a few boxes of enterprise drives and send them my way.
 
Last edited: