TR Forums

hbarnwheeler · Wed Jun 25, 2014 9:37 am

I was reading the latest update to the SSD endurance test here and thought I'd check in on my own drive, a Samsung 830 that's just over a year and a half old. To my surprise, Samsung magician reports that I've written 120TB to it! That seems crazy high to me (average 225GB/day). It's also well above some total write endurance specifications for new drives (the MX100 is rated at 72TB, for instance).

I turned off Windows indexing on that drive after installation, so I've no idea how this could have happened. I store all my media on traditional HDDs and use the Samsung as a system drive. Daily use is just basically web browsing and word processing, so I'm stumped.

All the relevant SMART attributes seem OK (at least for now), so I'm not worried about imminent failure, though I would like to figure out how this happened. Can somebody recommend an application I can download to monitor and log disk activity and track down the source of the issue? Given the price of SSDs these days, I'll probably replace it soon, though I'd still like to figure out how this happened. I can't post URLs, but relevant screenshots can be found here:

imgur-dot-com/a/XJb64

Wed Jun 25, 2014 10:02 am

How much RAM do you have, and is your pagefile on the SSD?

Do you have hibernation enabled?

derFunkenstein · Wed Jun 25, 2014 10:11 am

I might be totally wrong on this but looking at your screenshots it says your "OS is optimized for HDD". Could this mean that TRIM is not enabled? Does that have an impact on the number of writes? I'm curious what it says the optimizations would do.

I guess the other question is, does Magician support hte 830? Is it possible it's reading wrong?

Klyith · Wed Jun 25, 2014 10:37 am

derFunkenstein wrote:
I might be totally wrong on this but looking at your screenshots it says your "OS is optimized for HDD". Could this mean that TRIM is not enabled? Does that have an impact on the number of writes? I'm curious what it says the optimizations would do.

No, "OS Optimization" in the Samsung magician software is a terrible feature that does some really dumb "tweaks" -- for example setting the processor to run at 100% minimum in windows' power management options. No one should ever use it if they have a samsung drive, and after magician's performance in the SSD endurance test I'm skeptical of it's health indicators too. Though RAPID works much better than TR's review OTOH.

Trim can definitely have an impact on number of block written -- if trim is disabled the drive accumulates "dirty" blocks and these make write amplification happen as multiple blocks can get re-written for each original write. On non-sandforce drives trim is super important. The easiest way to check in W7 & up is to look in disk defragmenter and check that the SSD drive is greyed out. That means the os detected it as a ssd and should have auto-enabled trim and other ssd features.

hbarnwheeler you should get CrystalDiskInfo and double check with that.
http://crystalmark.info/software/Crysta ... dex-e.html

hbarnwheeler · Wed Jun 25, 2014 10:39 am

just brew it! wrote:
How much RAM do you have, and is your pagefile on the SSD?

Do you have hibernation enabled?

I've got 16GB of RAM, and the pagefile is on the SSD.

Hibernate is disabled, though "hybrid sleep" is enabled. Some googling suggests that this uses HD space, though even so, that doesn't seem to explain such a high number of writes, I wouldn't think.

hbarnwheeler · Wed Jun 25, 2014 10:47 am

derFunkenstein wrote:
I might be totally wrong on this but looking at your screenshots it says your "OS is optimized for HDD". Could this mean that TRIM is not enabled? Does that have an impact on the number of writes? I'm curious what it says the optimizations would do.

TRIM is enabled.

Klyith wrote:
No, "OS Optimization" in the Samsung magician software is a terrible feature that does some really dumb "tweaks" -- for example setting the processor to run at 100% minimum in windows' power management options. No one should ever use it if they have a samsung drive, and after magician's performance in the SSD endurance test I'm skeptical of it's health indicators too. Though RAPID works much better than TR's review OTOH.

Yeah, I didn't like what the SSD optimization wanted to do, so I didn't do it.

Klyith wrote:
Trim can definitely have an impact on number of block written -- if trim is disabled the drive accumulates "dirty" blocks and these make write amplification happen as multiple blocks can get re-written for each original write. On non-sandforce drives trim is super important. The easiest way to check in W7 & up is to look in disk defragmenter and check that the SSD drive is greyed out. That means the os detected it as a ssd and should have auto-enabled trim and other ssd features.l

Disk Defrag is set to "Never Run" on the SSD.

Klyith wrote:
hbarnwheeler you should get CrystalDiskInfo and double check with that.

I've got CrystalDisk info. It reports the same smart data as Samsung's Magician.

Thanks for your suggestions.

LaChupacabra · Wed Jun 25, 2014 10:55 am

That's a pretty massive amount of data. So much so it's hard to believe. A good place to start would be looking in Task Manager to see what your disk activity really looks like. You could even get fancy with it and setup some Performance Monitors to check every few minutes what state your disk is in.

Even without that are you doing anything like SETI or folding in the background? Have you run a virus scan lately? Hosting any media files or torrenting Linux distros?

Wed Jun 25, 2014 10:59 am

Hybrid Sleep is almost certainly the issue. If I am understanding correctly how Hybrid Sleep works, you are writing 16GB of data to the SSD every time the system sleeps.

Bauxite · Wed Jun 25, 2014 11:21 am

~7500 sleeps over 18 months doesn't seem plausible. Neither does torrents as ~7TB/month would get letters from many cable/dsl ISPs and most people use a platter drive for those folders

Seeding wouldn't cause any writes at all either, only fresh DLs.

This definitely sounds like windows and/or some program being stupid or malicious.

FWIW many torrent clients are quite good about buffers, file staging and allocation, they don't amplify writes hardly at all. I actually use an ssd for my 24/7 box as the only drive, but I have it set to copy and flush unseeded when it mounts the SAN share.

UberGerbil · Wed Jun 25, 2014 11:24 am

I have an 830 (120GB) that went into a new build as a system drive a little over two years ago. It's almost exactly the same setup -- media and data on another HD, 16GB of RAM with a hiberfile and a pagefile (that gets used a fair bit) -- and I left indexing (and Superfetch) turned on. Magician reports 11.39 TB written. That's an order of magnitude less, in almost twice as much time. Something is definitely wrong -- either the SMART data is messed up, or something is definitely writing way more data than a normal system should.

just brew it! wrote:
Hybrid Sleep is almost certainly the issue. If I am understanding correctly how Hybrid Sleep works, you are writing 16GB of data to the SSD every time the system sleeps.

Hmmm, let's do a little rough math. Assuming a full 16GB is written each time (I don't think that's the case unless the system is fully committed so the RAM is actually in use -- note that you can adjust the Hiberfile to be as little as 50% the size of your RAM), and ignoring the Gibi/Giga thing, 120TB of writes means that 16GB was written 7500 times. If the system has been running 20 months, that's 375 sleep events a month, which is more than 10 a day. Which I guess is possible if sleep is on a relatively short timeout. (Is that math right? I'm not through my first cup of coffee yet).

I just checked my system, and Hybrid Sleep is enabled (sleep is on a 2 hour timeout). Though it often doesn't sleep for various reasons. Still, I haven't seen anywhere near that kind of write traffic.

The Egg · Wed Jun 25, 2014 11:25 am

That just doesn't sound write.

*rimshot*

hbarnwheeler · Wed Jun 25, 2014 11:39 am

Thanks for all the replies guys. I agree with those saying that hybrid sleep doesn't really explain it all (though it's definitely a contributor). Virus scan came up clean as of a couple months ago, though I'll do it again. I just fired up the resource monitor and found something interesting.

I've been using the new version of Opera for awhile, and when it's running I'm seeing a total of about 10MB/sec disk activity, but when it's not running that drops down to 110k/sec (see graph at top right of screenshots). Also, the write B/sec under disk activity tends to really jump when opera is running. The main culprit seems to be the session backup manager.

The first screenshot is taken with Opera not running, the second while Opera is running. Still, however, this doesn't seem to explain the 120TB written. I've only been using Opera a few months, I'd say.

imgur-dot-com/a/4Balt

Wed Jun 25, 2014 11:45 am

Guess my math was a little off. :oops:

I still think it is possible that a combination of Hybrid Sleep and some VM paging could be responsible though.

Edit: Do sleep/hibernate events get written to the system event log?

SuperSpy · Wed Jun 25, 2014 12:04 pm

That might be possible if something is screwing with the machine's ability to stay asleep, like buggy wake on LAN or an overly-sensitive mouse.

Wed Jun 25, 2014 12:31 pm

SuperSpy wrote:
That might be possible if something is screwing with the machine's ability to stay asleep, like buggy wake on LAN or an overly-sensitive mouse.

Yeah, that's what I had in mind when I asked if sleep/hibernate events are logged.

Usacomp2k3 · Wed Jun 25, 2014 12:33 pm

just brew it! wrote:
SuperSpy wrote:
That might be possible if something is screwing with the machine's ability to stay asleep, like buggy wake on LAN or an overly-sensitive mouse.

Yeah, that's what I had in mind when I asked if sleep/hibernate events are logged.

Source: "Power-troubleshooter" Event ID 1 is what logs the wake up/resume.
I've got logs on my work laptop for "Wake source: S4 Doze to Hybernate" and "wake source: power button". I believe the former was a hybrid sleep initiative.

hbarnwheeler · Wed Jun 25, 2014 12:37 pm

Usacomp2k3 wrote:
just brew it! wrote:
SuperSpy wrote:
That might be possible if something is screwing with the machine's ability to stay asleep, like buggy wake on LAN or an overly-sensitive mouse.

Yeah, that's what I had in mind when I asked if sleep/hibernate events are logged.

Source: "Power-troubleshooter" Event ID 1 is what logs the wake up/resume.
I've got logs on my work laptop for "Wake source: S4 Doze to Hybernate" and "wake source: power button". I believe the former was a hybrid sleep initiative.

Thanks. I've got "Power-troubleshooter" log entries fewer than once a day going back months.

hbarnwheeler · Wed Jun 25, 2014 12:40 pm

Again, I don't know if it's enough to explain the 120TB total, but the new Opera really thrashes the disk. I set ProcMon to filter disk writes and I found this:

imgur-dot-com/ltWJONr

derFunkenstein · Wed Jun 25, 2014 12:47 pm

holy crap. I'd be inclined to switch browsers. That's unacceptable. Once you add hybrid sleep (if it goes to that mode twice a day and writes 16GB, that's 700+ 16GB batches per year, which is around 1.3TB) and installing apps and whatever virtual memory is used, you might have your total.

Wed Jun 25, 2014 1:01 pm

hbarnwheeler wrote:
Again, I don't know if it's enough to explain the 120TB total, but the new Opera really thrashes the disk. I set ProcMon to filter disk writes and I found this:

imgur-dot-com/ltWJONr

There's no way all of that is making it to the physical SSD, it must be mostly sitting in file cache. That's over a gig per second!

But if even a fraction of that is making it out to the drive, yeah that could certainly explain the 120TB, depending on how long this has been going on.

Duct Tape Dude · Wed Jun 25, 2014 1:22 pm

just brew it! wrote:
hbarnwheeler wrote:
Again, I don't know if it's enough to explain the 120TB total, but the new Opera really thrashes the disk. I set ProcMon to filter disk writes and I found this:

imgur-dot-com/ltWJONr

There's no way all of that is making it to the physical SSD, it must be mostly sitting in file cache. That's over a gig per second!

But if even a fraction of that is making it out to the drive, yeah that could certainly explain the 120TB, depending on how long this has been going on.

Wouldn't all of it make it to the SSD? Unless OP has explicity enabled a write cache, I'd bet the cache is read only and write through (instead of write back).

hbarnwheeler · Wed Jun 25, 2014 1:26 pm

just brew it! wrote:
There's no way all of that is making it to the physical SSD, it must be mostly sitting in file cache. That's over a gig per second!

But if even a fraction of that is making it out to the drive, yeah that could certainly explain the 120TB, depending on how long this has been going on.

Hmm, good point. I don't know much (read: anything) about how Windows file caching works. Are you suggesting that maybe it sits in memory, gets overwritten frequently, and then gets flushed to disk every so often?

FWIW: I get similar results when I leave Chrome sitting idle.

hbarnwheeler · Wed Jun 25, 2014 1:45 pm

Well, this is strange. I previously reported that indexing was off, and I got this information from "Indexing Options" under Control Panel. However, when I view my drive properties in Explorer, "Allow files on this drive to have contexts indexed in addition to file properties" was checked. Strange.

Edit: Some say this may be checked even if you've disabled indexing on the drive through control panel. I love Windows.

Klyith · Wed Jun 25, 2014 2:12 pm

Not every writefile in process monitor will necessarily hitting the disk. You can use diskmon from sysinternals to watch what bits are hitting the disk, which are frequently much smaller than those in procmon.

hbarnwheeler wrote:
FWIW: I get similar results when I leave Chrome sitting idle.

So when you were looking at Opera, was that stream of writes constant like that screenshot shows, for long periods of time? Chrome & Firefox on my system will also produce frequent blobs of writes for session info, but it happens in chunks every 5-15 seconds. If they are idle, it takes more like a full minute to accumulate a similar size list of writes to what you got in 1 second.

Windows indexing shouldn't produce a whole lot of total writes, so unless something is wrong I'm not sure you need to worry about it. One question: was your windows install originally on a HD and get cloned to the SSD, or did you start with a fresh install?

I actually took Chrome's install / user directory thing and shoved it on a standard HD via a NTFS junction link, just because Chrome was crapping up my SSD with its obnoxiously huge cache. I only have a 90g drive, you don't need to cache 5 gigs of youtube videos! But that was just a space issue.

hbarnwheeler · Wed Jun 25, 2014 2:42 pm

Klyith wrote:
So when you were looking at Opera, was that stream of writes constant like that screenshot shows, for long periods of time? Chrome & Firefox on my system will also produce frequent blobs of writes for session info, but it happens in chunks every 5-15 seconds. If they are idle, it takes more like a full minute to accumulate a similar size list of writes to what you got in 1 second.

It was pretty constant, I think, though every 5 seconds seems to qualify as "constant".

Klyith wrote:
One question: was your windows install originally on a HD and get cloned to the SSD, or did you start with a fresh install?

I believe it was a fresh install.

The Egg · Wed Jun 25, 2014 2:51 pm

I dunno, roughly a month ago I installed a cheap 840 EVO to use as a cache drive. That's Windows swapfile, Windows temp directory, browser cache, photoshop scratch disk, etc, etc. After a month of use, I'm sitting at 0.05TB, or roughly 50GB of writes. Consistantly pulling off 224/GB of writes each day would be almost impossibly difficult without a program designed to do so.

Wed Jun 25, 2014 3:03 pm

just brew it! wrote:
Hybrid Sleep is almost certainly the issue. If I am understanding correctly how Hybrid Sleep works, you are writing 16GB of data to the SSD every time the system sleeps.

No it doesn't. It at most uses 75% of the total RAM capacity and can even be forced down to 50% and those are just the sizes of the file. It does not use that entire capacity.

Also the Pagefile barely generates any writes at all.

Wed Jun 25, 2014 3:08 pm

Link

Microsoft wrote:
Should the pagefile be placed on SSDs?

Yes. Most pagefile operations are small random reads or larger sequential writes, both of which are types of operations that SSDs handle well.

In looking at telemetry data from thousands of traces and focusing on pagefile reads and writes, we find that
•Pagefile.sys reads outnumber pagefile.sys writes by about 40 to 1,
•Pagefile.sys read sizes are typically quite small, with 67% less than or equal to 4 KB, and 88% less than 16 KB.
•Pagefile.sys writes are relatively large, with 62% greater than or equal to 128 KB and 45% being exactly 1 MB in size.

In fact, given typical pagefile reference patterns and the favorable performance characteristics SSDs have on those patterns, there are few files better than the pagefile to place on an SSD.

Are there any concerns regarding the Hibernate file and SSDs?

No, hiberfile.sys is written to and read from sequentially and in large chunks, and thus can be placed on either HDDs or SSDs.

Wed Jun 25, 2014 3:33 pm

Ryu Connor wrote:
just brew it! wrote:
Hybrid Sleep is almost certainly the issue. If I am understanding correctly how Hybrid Sleep works, you are writing 16GB of data to the SSD every time the system sleeps.

No it doesn't. It at most uses 75% of the total RAM capacity and can even be forced down to 50% and those are just the sizes of the file. It does not use that entire capacity.

Also the Pagefile barely generates any writes at all.

Yeah, I think we've already moved on from that theory. As noted a few posts back, it looks like Opera may be the culprit.

hbarnwheeler · Wed Jun 25, 2014 4:20 pm

just brew it! wrote:
Yeah, I think we've already moved on from that theory. As noted a few posts back, it looks like Opera may be the culprit.

I hope so (if only because I would know for sure). But, like you said, most of that may just be cached and never see the disk.

In any event, it looks like I'll be shopping around for a replacement.

Amazon's got the EVO for $135 today. I'd look at the MX100, but it doesn't come with any software and TR has reported that its SMART attributes are rather cryptic.

TR Forums

120TB(!) writes on relatively young Samsung 830. What's up?

120TB(!) writes on relatively young Samsung 830. What's up?

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Re: 120TB(!) writes on relatively young Samsung 830. What's

Who is online