Is 2TB/mo too much to write to an SSD?

[VirtualLarry]

It seems my BOINC client writes about 2TB/mo to my SSD. It says it has been working for 4 months, and it's already down to 76% lifespan, with nearly 8TB written.

 

[Golgatha]

I moved my Folding@home clients to a mechanical drive, although that was more for making backups easier for my C: drive. Interesting question.

 

[corkyg]

SSD's should not be accessible to auto on line downloads. That's what spindles are for.

 

[wpcoe]

Ack. What? You mean Folding@Home is responsible for all that SSD activity?

I was surprised to see that my SSD has 1.82TB reads/1.75TB writes in 533 hours. I knew I ran a ridiculous amount of benchmarks testing various configurations of Intel/Marvell/Asmedia SATA ports, but is it really F@H that's the culprit?

I assume it's just a matter of uninstalling it and re-installing it on a HDD?

 

[bryanW1995]

http://www.anandtech.com/Show/Index...intels-3rd-generation-x25m-ssd-specs-revealed

I would not run F&H on that ssd anymore. 8tb is getting into no man's land on an x25m g2, and iirc you're on a vertex. If you get 30+ TB out of your ssd I'd be impressed.

 

[Anteaus]

I don't much about SSDs but this thread interested me. Are you guys saying that SSDs start dying from day one based on how much data is written to it? I probably misunderstood what you guys were alluding to but that's what it sounds like to me. Thanks for the info.

 

[exdeath]

It seems my BOINC client writes about 2TB/mo to my SSD. It says it has been working for 4 months, and it's already down to 76% lifespan, with nearly 8TB written.

What do you have, a 4 GB SSD?

If one of my SSDs had 8 TB written it would be at 99.69% life.

Seems you have a low capacity drive that is really old with poor write amplification and wear leveling or just uses piss poor NAND. A SF2281 with 32nm NAND would last (2560 TB total write capacity) / (2 TB mo) = 1,280 months = 106 years.

 

[frostedflakes]

67GB/day from a DC client? That's nuts, about the same kind of load the database server that runs the AnandTech main site receives.

http://www.anandtech.com/show/5518/a-look-at-enterprise-performance-of-intel-ssds/1

Those kind of writes will chew through an MLC SSD in no time, consumer SSDs just aren't designed to handle that amount of writes. Move that program over to an HDD.

 

[ViviTheMage]

what is a good prgram to check these stats? I have OCZ drives mostly.

 

[Joepublic2]

I don't much about SSDs but this thread interested me. Are you guys saying that SSDs start dying from day one based on how much data is written to it? I probably misunderstood what you guys were alluding to but that's what it sounds like to me. Thanks for the info.

Yeah they do, it's a limitation of the technology. Didn't realize it happened that fast with folding@home, though, I'll be sure to avoid running it on a SSD. I doubt it's the total volume of data that's killing it though, it's more likely bad wear leveling algorithms combined with lots of random writes from the client.

 

[cantholdanymore]

My Intel 510 128GB shows 100% after 3.7TB writes

 

[LOL_Wut_Axel]

It seems my BOINC client writes about 2TB/mo to my SSD. It says it has been working for 4 months, and it's already down to 76% lifespan, with nearly 8TB written.

Where and how do you check this?

Now I'm worried because I run folding@home on my Vertex 3.

 

[gevorg]

To reduce stress on SSD drive and increase its life, use RAM-disk for heavy use folders like browser cache. Some RAM-disk utilities even let you image and save the content of its drive before reboot.

With 16GB system memory, I've allocated 4GB RAM-disk for cache of all my browsers, as well as temp folders for some programs/utilities I use.

 

[VirtualLarry]

I changed my task checkpoint interval in BOINC to 600 seconds, rather than 60 seconds.

 

[VirtualLarry]

What do you have, a 4 GB SSD?

If one of my SSDs had 8 TB written it would be at 99.69% life.

Seems you have a low capacity drive that is really old with poor write amplification and wear leveling or just uses piss poor NAND. A SF2281 with 32nm NAND would last (2560 TB total write capacity) / (2 TB mo) = 1,280 months = 106 years.

OCZ Agility 30GB, with 1.7 firmware.

 

[frostedflakes]

To reduce stress on SSD drive and increase its life, use RAM-disk for heavy use folders like browser cache. Some RAM-disk utilities even let you image and save the content of its drive before reboot.

With 16GB system memory, I've allocated 4GB RAM-disk for cache of all my browsers, as well as temp folders for some programs/utilities I use.

I never really worried about page file, browser cache, etc., none of these write very much to the disk. Other than putting movies, music, stuff like that on an HDD (mostly for space reasons), I don't really do anything to avoid writing to my SSD other than avoiding the obvious and unnecessary stuff, like defragmenting, excessive benchmarking, etc. Page file is set to the default (8GB since I have 8GB of RAM), browser cache is on the SSD, I install/uninstall Steam games to the SSD fairly often, I reinstall Windows probably once every 2-3 months on average, I recently installed Windows 8 to the SSD to try out before deciding I didn't really like it after only a day or two and restoring to my Windows 7 backup image, etc. Basically I don't go to great lengths to avoid writes and I've still only averaged about 11GB/day of writes over the last year and four months or so. Just to put 67GB/day of writes into perspective, that is *a lot* and would worry me with a smaller (32-40GB) SSD and/or one with an older controller that doesn't have very low write amplification. As exdeath pointed out it's probably not as much of an issue for larger SSDs with good controllers, though. For example the 120GB Corsair Force 3 that was endurance tested by XtremeSystems members was able to handle about 430TB before MWI was exhausted and didn't actually die until 1PB of writes. Assuming 2TB/mo that's still like 18 years before you'd exhaust MWI. Would still try to move those writes over to an HDD since they probably don't need the speed SSD offers and to me would just seem to be unnecessarily wearing it out, though.

 

[Emulex]

remember 2TB spaced out is not the same as 2TB in large chunks due to how sandforce works. plus the slower the nand is written - the longer it will last.

why the intel 710 is just the same as 320 with 50-60% overprovision and firmware that drastically slows down Program/Erase (PE) speed.

 

[Joepublic2]

remember 2TB spaced out is not the same as 2TB in large chunks due to how sandforce works. plus the slower the nand is written - the longer it will last.

why the intel 710 is just the same as 320 with 50-60% overprovision and firmware that drastically slows down Program/Erase (PE) speed.

Yeah, I wish more manufacturers would be more conservative with the write speed because typical desktop usage patterns are about 95% reads and 5% writes (if that).

 

[VirtualLarry]

Pics!

 

[wpcoe]

Until further notice, I've stopped running F@H. I thought I was being altruistic by leaving my computer on 24/7 and running it.

Googling about moving F@H goes into alien discussions about moving SMB directories, and apparently there are different ways to download, install and implement F@H, and I'm just not motivated to learn yet another arcane obscure black art.

 

[Mr. Pedantic]

FAH isn't like BOINC, it should only have one read burst and one write burst per WU, as it's loading data into memory and writing it back to disk. BOINC and several other DC projects are really bad, though.