Any problem with 2TB WD Cavier Green (WD20EARX) and Freenas?

[isaac82]

Hi,

I'm replacing a ReadyNAS Duo with the following:

1 x HP ProLiant Turion II N40L MicroServer
1 x Kingston 8GB (2x4GB) DDR3 1333MHz i5 Memory Kit Non-ECC CL9 1.5V
4 x WD 2TB 3.5" SATA-III 6GB/s Caviar Green Hard Drive - 64MB Cache - WD20EARX (for RaidZ)
1 x Transcend JetFlash 700 8GB USB 3.0 Flash Drive (for the OS)

A friend made me aware of a problem he encountered a while back with WD Green discs and Linux, where the park head idle timeout was ~5s and Linux does some write operation every 10 or 30s and this results in a constant park/unpark situation, with a high SMART LCC as an indicator.

I've tried searching this forum and others for whether this is/was a problem with FreeBSD/FreeNAS and haven't found anything conclusive.

My question is simply whether anyone can tell me if this is a problem with FreeNAS 8 and this the WD20EARX?

Thanks for your time.

 

[cyberjock]

It's not a problem with linux, FreeBSD, FreeNAS, DOS, Windows, etc.

It is a problem(aka feature) of the drive itself, regardless of the OS. I have 22 WD green drives in a FreeNAS server. All you have to do is make sure to read up and use the Western Digital tool 'wdidle' from Western Digital to change it to whatever you want. I set my drives to the maximum (300 seconds) but you can also disable the feature completely.

 

[isaac82]

Thanks for the reply. I asked if it was a problem with FreeNAS as in that disk + freenas/freeBSD = problem, as that feature of the drive on it's own isn't a problem. It only becomes a problem when combined with a OS that frequently reads/writes to the disk. Like Linux does and apparently windows doesn't.

All the forum and news articles that relate to it are at least a year old, so I wanted to also check that the wdidle tool is still needed and still works on newer drives.

Also, I believe that wdidle tool is a CD iso boot image to dos. I don't have a CD drive, does anyone know if you can put that image on a USB stick?

 

[cyberjock]

Thanks for the reply. I asked if it was a problem with FreeNAS as in that disk + freenas/freeBSD = problem, as that feature of the drive on it's own isn't a problem. It only becomes a problem when combined with a OS that frequently reads/writes to the disk. Like Linux does and apparently windows doesn't.

You are still misunderstanding the "problem". It has nothing to do with the OS at all(which is what I thought I made completely clear previously). It has to do with the type of disk writes you make. I'm going to go on a limb here and say that Linux doesn't write to a drive every 30 seconds or so without a reason to. If you have a file share then one of your clients likely is writing. You may have some disk utility or something. If you were to put the same workload as that Linux machine on a FreeNAS server, you will have the same problem. If you did it on Windows, DOS, or any other OS, as long as you had the same loading for the file shares or whatever was initiating the writes, you WILL still have the same problem. If you ran a database on your server, you could potentially expect the same problem. It's about what the drives will be doing and not what OS you are on. You need to do alot of reading up on what the wdidle tool is and how it fixes the potential problem. Some servers will never have a problem even with the default settings. For the longevity of your drives, you should go read up on this "feature" until you understand it. I hashed out a very long post in this forum somewhere for Green drives. If you want to you can go read that. I don't want to spend another hour retyping all of that.

So.. I'll repeat myself: It's not a problem with linux, FreeBSD, FreeNAS, DOS, Windows, etc.

All the forum and news articles that relate to it are at least a year old, so I wanted to also check that the wdidle tool is still needed and still works on newer drives.

You'd have to check out the WD forums for that answer. It's WD's baby. But I'm sure they could make a small firmware change and that feature would be gone. From what I've read, that has happened in the past, but was later reimplemented.

Also, I believe that wdidle tool is a CD iso boot image to dos. I don't have a CD drive, does anyone know if you can put that image on a USB stick?

Again, you'd have to check out the WD forums for that answer. It's WDs baby.

 

[isaac82]

I'm not sure if we disagree on the issue or if we're just using different terminology. My understanding is:

WD disks have a short timeout to park the heads. This isn't a problem in itself, but a design choice to save energy. It becomes a problem if those disks are used in a system (let's say for arguments sake by that I mean everything running: OS+Services+installed apps) that reads or writes to the disks on a short frequency. In this situation the disk goes idle, parks then the system wakes it up and unparks. Repeat until you've worn out the disk.

So I think the question is valid on whether an install of freeNAS triggers that scenario. You say you don't think Linux would write every 30 seconds or so, but again from what I've read, that is the case thanks to the system logs it writes. The question is does freeNAS write logs frequently? I've not problem with occaisional user traffic causes a few parks and unparks, but if the system logs or does something else to wake the discs that will be a problem.

Anyway, something else that just occurred to me, if it's the OS logging that causes the periodic wakeup that conflicts with the short park time - is this still a problem if I'm running freeNAS from a USB stick and just use the 4 WD Green drives for storage? I assume all OS logs are in one of the two USB OS partitions?

BTW I am also on the WD site (have emailed support and posted on the forum a question about whether the wdidle program supports my drive model). Also, I'm not accusing FreeNAS of having a bug or anything, just need to know if the design choices of these drives and FreeNAS cause the problem scenario.

 

[cyberjock]

I'm not sure if we disagree on the issue or if we're just using different terminology. My understanding is:

WD disks have a short timeout to park the heads. This isn't a problem in itself, but a design choice to save energy. It becomes a problem if those disks are used in a system (let's say for arguments sake by that I mean everything running: OS+Services+installed apps) that reads or writes to the disks on a short frequency. In this situation the disk goes idle, parks then the system wakes it up and unparks. Repeat until you've worn out the disk.

So I think the question is valid on whether an install of freeNAS triggers that scenario. You say you don't think Linux would write every 30 seconds or so, but again from what I've read, that is the case thanks to the system logs it writes. The question is does freeNAS write logs frequently? I've not problem with occaisional user traffic causes a few parks and unparks, but if the system logs or does something else to wake the discs that will be a problem.

Anyway, something else that just occurred to me, if it's the OS logging that causes the periodic wakeup that conflicts with the short park time - is this still a problem if I'm running freeNAS from a USB stick and just use the 4 WD Green drives for storage? I assume all OS logs are in one of the two USB OS partitions?

BTW I am also on the WD site (have emailed support and posted on the forum a question about whether the wdidle program supports my drive model). Also, I'm not accusing FreeNAS of having a bug or anything, just need to know if the design choices of these drives and FreeNAS cause the problem scenario.

You say "potato' and I say "potaahto'.

The OS should only be writing or reading from the disk if it is requested to do some for some reason besides "it's linux" "its windows' etc. Your friend had something either running on the server, something using a file share, a database, or something requesting those frequent reads. If I bootup any properly setup server and let it sit there I should NOT be seeing disk activity. This is true for virtually all operating systems. The system should sit there wasting electricity doing absolutely nothing productive. FreeNAS(or linux, or windows, or any other OS you want) will bootup, do some bootup sequences with your zpool/ufs/whatever and then sit there idly by waiting for input from a client, another program or service on the server, or something.

What DOES matter is what your system will be used for. If you plan to run huge database with 1000s of transactions an hour, you'd BETTER have that wdidle tool setup and working. If you plan to share your home pictures and video, then it's much less important.

What really matters is the loading of your server. EVERY server should bootup and sit there. If its not its because the administrator did something to make it not sit there, either by installing something on the server or using services to share files/data.

I wrote in the previous post, but deleted it, but your friend isn't as knowledgable as he thinks if he was blaming "linux". The reality of the issue with Green drives is poor administration of hardware he is not knowledgable in. It is an obscure issue, but anyone sufficiently thorough in their evaluation and choosing of hardware should have known about this issue, read up on it, and understood the problem. He'd also know that if the drive has a small number of hours recorded as online time in the firmware and you try to RMA the disk because of a SMART imminent failure because of disk parks WD will NOT issue an RMA. You are simply screwed. It's a dead givaway that you weren't using the drive in a desktop(which is the only thing the drive is designed for). Your warranty is void because you used the drive outside the "designed" use of the drive.

At the same time, these green drives work pretty darn well in desktops. Despite the millions of people using them in linux, Windows, etc daily, the drives perform well without the problem of the green features. The reason is because the OS doesn't do reads/writes every 30 seconds like your friend claims, but also your average desktop user won't be running databases, sharing TBs of data cross a LAN, and generally doing things that result in constant read/write requests.

For the record, the head parking feature is officially called "intellipark". I had to look it up since I couldn't remember the name.

You need to forget that your friend blamed "linux" for his problems and start understanding that the limitation is on how frequently reads and writes are requested.

Edit: Grr.. clicked post instead of preview.

If your system is writing system logs every 30 seconds, then something is broken with the system and the admin should be fired. The whole point of logs is to identify problems and see when they started. If you are making new entries every 30 seconds it won't take long before your logs are only a day or 2 long. That defeats the purpose of the log. Not to mention the performance penalty with having to constantly seek the heads to write a 1kb log entry. You shouldn't be accumulating tons of logs like that. In any case, FreeNAS logs are saved to the ramdisk and discarded on reboot. So logs won't be the problem.

I have 22 green drives in my FreeNAS server, been in Windows server and FreeNAS server for a combined 3 years. Never had a drive fail until last night(grr). 16xWD20EARS and 6xWD30EZRX. All drives have performed well. Now that you've made me talk about this for 45 mins or so, I'm curious to find out how many parks are logged in the SMART data.

 

[cyberjock]

All of my drives consequently have less than 5000 cycles on SMART attribute 193. Ironically, the drive that failed last night has 260k cycles. So it appears that I somehow missed a drive on accident.

 

[isaac82]

I thought I'd have a look into where this claim "Linux" triggers the issue came from.

From a google, this looks like an early thread on the issue: http://www.silentpcreview.com/forums/viewtopic.php?f=7&t=51401

One post in particular is interesting, unfortunately the link to the mailinglist is broken, so I can't be sure what the conclusion of the discussion was:

Apparently Linux tries to save the discs by buffering data and syncing it to discs every 10-15 seconds. Unfortunately this is enough time for the WD to park it's heads..

Here is pointer to discussion about the issue on Linux Kernel Mailing List: http://kerneltrap.org/mailarchive/linux ... /9/1386304

Disclaimer: I am not expert in Linux kernel, so my understanding about how the Linux kernel works with discs is pretty shallow. The only thing I know for sure is all the 4-6 GPs in my or my friends possession have alarmingly high head load/unload counts.

If the Linux kernel buffers writes just longer than the park time, that would explain the issue.

Anyway going back to my original question, which was effectively 'do FreeNAS and WD Green play nice?', you've made it quite clear you think only a poorly administered system would make frequent read/writes on it's own, so I'll take it a clearn install of FreeNAS will be fine.

It only leaves the slightly contradictory statement you made originally where you use FreeNAS but always set your WD Greens to a 300 second idle.

 

[cyberjock]

If the Linux kernel buffers writes just longer than the park time, that would explain the issue.

You are also making the assumption that there ARE writes that need to be buffered. My whole argument is that you shouldn't have a constant stream of unknown writes or reads from the OS. Buffered, unbuffered, cache in memory, cache on a RAID controller, its a totally different discussion. If no writes are being performed, then absolutely nothing matters regarding the size of the cache, the type of cache, etc. That's why I say only a poorly administered server would have "random" writes for no reason. For ZFS, it does its write caching in a very unique way. Each OS, each hardware design will handle write caching different for write-through as well as write-back. I'm sure neither of us has $25k+ to go buy a bunch of different types of hardware to run all sorts of tests to figure out which one for a given type of load will optimize writes against WD Green drives.

Anyway going back to my original question, which was effectively 'do FreeNAS and WD Green play nice?', you've made it quite clear you think only a poorly administered system would make frequent read/writes on it's own, so I'll take it a clearn install of FreeNAS will be fine.

It only leaves the slightly contradictory statement you made originally where you use FreeNAS but always set your WD Greens to a 300 second idle.

I didn't say that. I said that a poorly administered system would make frequent read/writes with no known cause(and no.. you can't blame the "OS"). If I move 1TB of data to my server from a workstation, it had better make frequent writes. If you know what kind of loading your server plans to have, you should be able to optimize your idle setting appropriately. If you don't even know what the idle setting is, you have a problem. If you have a service running that's supposed to be causing those reads and writes, that's fine. It's when you are oblivious to what's going on and blaming things that ultimately can't be responsible for the cause that things get messy. Do you blame guns for killing people? Do you blame cars for killing people? It's almost the same thing.


I've seen SQL servers that over a period of months they began to perform very poorly. A SQL admin was called in, and in about 3 days of understanding what data our database stores and a few tweaks we achieved almost 2500% increase in performance. We also found that the hard drives storing the database weren't failing as quickly as they had been. Instead of replacing a drive every 1-3 months we were replacing 2 a year. Our SQL admin should have known better. It became very obvious he really didn't know what he was doing, and months later he was unemployed.

A good admin should know his hardware inside and out. He should know what needs to be optimized to maximize performance and reliability. If he's being blindsided by "features" a company is using as selling points, thats a VERY bad sign. If he's being blindsided by a firmware bug that's a VERY good sign(that means he knows his stuff and was able to figure out it was a firmware bug). In this particular case, it's not a firmware bug. It's a feature that doesn't work well in server environments and your friend didn't know what he was doing when he bought the drives.

Yes, if you setup WD greens with either 300seconds(that's the maximum besides disabled) I'd say that's a good choice. It's not necessarily the most "power savings" you can get, but the drive is still fairly low powered even when the head isn't parked.

This is a recurring issue with Linux and Unix. It is necessary for you to "know your shit" when working with these OSes. You can't ask a question and get an easy answer. You really have to know what you are doing and understand what enabling(or disabling) things does, deep down at times. Too many people don't want to do that, and this forum consistently gets asked the exact same questions over and over because people either don't want to read, didn't understand but didn't figure it out, or think they actually know what they are doing and they don't.

 

[vkshatriya]

N40L wiki says .. WD Green is not recommended. http://n40l.wikia.com/wiki/Hard_drives

I had some issues with WD Green, N40L and FreeNas with default settings. I seen timeout errors while writing some large files..

 

[isaac82]

'WD Green drives are not recommended at this present time due to firmware issues related to power down.' is a bit cryptic.

I've been running my setup for a couple of weeks now - not put as much time into learning it as I'd have liked due to lack of time, but it seems to be fairing ok so far.

I've been running smartctl -a /dev/ada... to get smart stats out (shame there's nothing in the web front end for this) and my LCC seems ok. Think it's approaching 200.

I still need to look into how and when to run the smart self tests - smartctl can output some interesting hints on how long the results should live - which I presume means how long before rerunning them. I need to do some googling to find out what tests people typically do. What I want is to add a decent self test on boot, with warnings emailed and maybe cron another set once a day. The system will basically go unused 99% of the time, so that should suffice.

My only real complaints with the system at the moment are NIC related. Seems the BSD drivers don't support the broadcom onboard NIC properly - wake on lan and jumbo packet support being the big losses. I could buy an intel NIC, but I'm reluctant to just to gain WOL and increased transfer speeds. Tempted to see if maybe better drivers come.

 

[cyberjock]

Well, I have a server with 22xWD Greens and it has had zero problems for 2 years.

I'm thinking it's not related to powerdown but related to the intelli-park that could potentially hit the SMART threshhold rather quickly in a RAID environment if not adjusted properly. Rumors on some websites is that some people were hitting the 300k SMART failure threshhold in less than 2 months in a RAID environment.

 

[isaac82]

The reply I got from WD support was not to try and run the idle3 firmware tool on the drive as it wasn't supported. Many posts on the net suggest it works, but I decided to just monitor the LCC value and only try and fix it if it looks to be increasing at an undesirable rate.

 

[geater1]

Hi,

I also have an N40L server which i have set up in the last 2 days. It has 2 GB or RAM and 3 x 2 TB WD green caviar hard drives. I have set up the system as a ZFS Raid-Z based on advice on forums.
I have been using the NAS to store video files for playback over the home network through a WDTVlive media player. So it gets used daily for an hour or so when i am watching videos.

The hard drive configuration settings in FreeNAS (e.g. idle time and SMART) have been left as standard with no idle and SMART enabled. Going through the forums i am concerned i may have made the wrong choice in hard drive. Are there any settings in FreeNAS which i should change based on my usage? Also in the FreeNAS reporting page is there anything i should keep an eye out for to indicate there is an issue with the drive.

Thanks for any and all help.

 

[isaac82]

geater1 I wouldn't worry too much, but I can suggest a few things to monitor.

Setup a user that you can ssh in with and make sure you can use su (think you have to be a member of wheel).
Run su then with root privileges run
smartctl -a /dev/ada0
This should output lots of SMART stuff (think the -a is all, you can refine if you're not interested in it all with different args).
The thing you probably want to keep an eye on is the Load Cycle Count (LCC). This is the worry with this hard disk as the intellipark feature tries to save a bit of power by parking the head after about 8 seconds. I think for a HDD the lifespan is something from a few hundred thousand to a million, so this number should and will go up, but it's the rate you want to keep an eye on.

BTW my brief instructions above are what I've got around to so far (time limited by wife/child/work/etc) but you might get better mileage out of this forum post:
http://forums.freenas.org/showthread.php?6211-Setup-SMART-Reporting-via-email

I'm also going to look into setting up a schedule for some self tests and enabling the disk standby time, I'll give some details here when I've done that. One useful thread you might want to look at is about seeing if your disks are going into standby or not (although you say you've not enabled this so they won't for you):
http://forums.freenas.org/showthread.php?2068-How-to-find-out-if-a-drive-is-spinning-down-properly

If you've set everything up as you say and just use it as a home NAS for your own video storage you should be fine. If you were to run a torrent client on there or some other app or plugin that frequently accesses the drive (but less frequently than say 8 seconds) you might need to change some settings.

BTW I'd say from what I've read that 2GB RAM is not sufficient for your setup (I'm guessing that's the 2GB ECC that came with it). I bought 8gb non-ecc for about £25 and I think that would be something worth looking into. ECC would be better if you can be bothered to find some and spend the extra.

One question for you, have you done any disk read/write speed tests? I used DD to check the speeds I'm getting and get just over 200MB a second - that's with raidz1 with 4 2tb WD greens. I'd be interested to know what you get with 3 disks as really that's a more optimal setup for raidz1.

Good thread here for how you check the r/w performance: http://forums.freenas.org/showthread.php?981-Notes-on-Performance-Benchmarks-and-Cache

My results:

Code:

Write:

[isaac@freenas] /mnt/volume1/media# dd if=/dev/zero of=tmp.dat bs=2048k count=25k
25600+0 records in
25600+0 records out
53687091200 bytes transferred in 231.315083 secs (232095074 bytes/sec)

Read:

[isaac@freenas] /mnt/volume1/media# dd if=tmp.dat of=/dev/null bs=2048k count=25k
25600+0 records in
25600+0 records out
53687091200 bytes transferred in 263.746570 secs (203555600 bytes/sec)

 

[geater1]

Thanks Isaac82, much appreciated. There are a fair few things you described which i will have to research to figure out how to do. I haven't done any performance tests yet. Have just been transferring files to the NAS through WiFi. Will re-post when i have the results.