Momentus XT in RAID 5

Over in the Clevo x7200 thread, we got off topic a bit with a user having problems with Momentus XTs in RAID 5 configuration.

I also posted something to the Seagate forums, and haven't heard back.

Anyone running XTs in a notebook w/ RAID 5? Care to share your experiences?

 

Generally when people have trouble with RAID it is a controller problem. Has that person tested RAID 5 with another drove model?

 

Probability, speculation on my part, is the drive is sleeping and not being writen too properly. It could be just them waking or just not getting written too. I believe Segates resonce is XT's are not raid capable drives period.

What may help actually is SD23. The reason I say this is supposedly it turns off the drivres ability to sleep and spin down at all. Of course this can cause other issues related to disabling power management. Before SD24 I do not think the srives were supposed too respond too OS power management at all and work off internal power management.

Another possability is a bad drive. XT's are infamous for hangups from a drive the runs out of conformance. if one of the three drives is doing this it could be causing an issue as well.

In the end XT's in Raid0 can be problem enough, complicate it by Raid5 and the problems can be multilied exponentially. The Sager is one of the few, if not only, laptops alowing Raid5 with 3 drives so you will probably be limited in finding others with the same issue................

 

I wouldn't be able to put my finger on exactly why, but using those hybrid drives in a RAID array just seems like a really really bad idea. The clever Momentus controller shuffling things about to the flash memory and the RAID controller itself aren't really working towards the same ends.

And how (and why?) are people doing RAID 5 in laptops?! It needs minimum three drives, and I've never seen a laptop with four drive bays.

 

It's because the performance advantage of the Momentus XT is wasted in RAID5.

RAID5 splits data up across the drives in (n-1) parts. 3 drives in a RAID-5 array means that 2 of those 3 contain data, with the 3rd drive containing parity / error checking information. In order for the Momentus XT speed benefits to be used, the data you are reading must be in the 4GB SLC NAND cache of all 3 drives. If the data is only cached in the SLC NAND flash of 1 out of 3 drives, you will be waiting for the mechanical read speed of the remaining 2 drives before the RAID controller serves up the data. You're essentially using the Momentus XT drives as 7200rpm drives.

On top of that, you typically do not use RAID5 arrays to store any kind of data with load times because of the performance hit you get. Things like OS, applications, games, etc (anything with a load time) goes on the fastest drive you have; on desktop systems, this would be a single fast mechanical drive or SSD; on servers, this would be a RAID-1 mirrored disk array. You don't put things with load time on RAID-5.

What you DO put on RAID-5 is bulk content. On desktop systems, you put media that gets consumed... videos, photos, documents, pr0n, etc. And for that kind of stuff, read speeds don't matter. A 1080p video will play back equivalently well on RAID-5, on bare drives, even on USB 2.0 interfaces.

 

Heh, cheers. If I'd thought about it properly I'd have come to the same conclusion about the contents of the flash portions needing to be exactly synchronised, but I try not to clutter my brain with too much DBA witchcraft if I can avoid it

Edit: tell you what really scares me though, is I looked up that Clevo model posted above, and it has three hard drive bays. So they're booting off a RAID5 array?!

 

We thought that as well. The user dropped in 3 Western Digital Scorpios and has been smooth sailing for the last 6+ days. When the XTs were in, one of the drives would go to degraded in about 4 days.

 

That is what I would like from Seagate, an official announcement. Of course, Mark Wojtasiak from Seagate says "No you will not have any problems with RAID 5. To the system the Momentus XT just looks like a hard drive." But he is not an engineer. I would like the straight dope from Seagate engineers / support. Unfortunately, Seagate support remains quiet.

No go there. We've been working over this problem over 3+ weeks. The problem started w/ drives on SD23, and then persisted when they were upgraded to SD24.

Possibly.

Agreed. I wonder if we'll ever find an answer. If anyone wants more background on the "Clevo x7200 and Sager NP7280 owners lounge" thread, search for wwoods, scook9, and my posts which cover this issue.

 

There's no shuffling of data here, just mirrored data in the SSD cache.

The Clevo x7200 has 3 bays, and if you choose, you can configure the machine w/ out an optical drive, use 4 drives and get RAID 0+1!

It is a sweet, sweet machine.

 

Actually, the cache of ALL three drives are used. Don't think of the cache as a FILE cache, but rather a disk LBA-addressed cache.

So when the OS asks for a file over a 3 disk array, the drivers know they need to get 1st block from A, 2nd block from B, and CRC from C, over and over. The drive itself merely stores copies of the most requested BLOCKs in the SLC NAND cache. So as a block is requested over and over, it ends up in the SSD cache. So in effect, you end up w/ 12GB of cached data with files hacked up into 2 pieces and the third used as a checksum.

In general use, but the x7200 is a desktop replacement w/ 3 drive bays. The user's choice of RAID 5 was not for performance reason, but more out of safety / uptime. He says he's adamant about backups (and I have no reason not to believe him), but having the RAID 5 in place allows you to keep a laptop up and running w/ minimal down time in case of a single disk failure, and RAID 5 would provide some better write times vs. the only other "safety" option of RAID 1.

 

Yes, I understand that it is a block-level cache, and not a file-level cache. I never made any reference to "file" in my post.

But the point remains - if the complete set of data that the disk controller is requesting isn't in the SLC NAND Flash memory of ALL of the Momentus XT drives, then you're still waiting on a mechanical 7200rpm drive to spin and find the requested data.

You may get very specific situations when you DO get your data in cache. But for every other instance when your data is NOT in cache, you will be using the Momentus XTs as regular 7200rpm spinning disks... which is not good, because you incur a performance hit by being in RAID5, and you incur a performnace hit because the Momentus XTs are slower than other 7200rpm drives when doing read operations off of the mechanical disk.

Meh.

I am a firm believer that just because you CAN do something, doesn't mean you should. I would have configured my system a little differently for max uptime, and not used RAID5 at all.

But to each his own.

 

Ah OK. That eases me worries about people booting off RAID5!

Still unconvinced it's useful for anything apart from servers/storage arrays though really.

 

Not sure that is possible. The most requested items end up in the cache based on an algorithm of LBA. When the OS asks for files over and over again, the blocks within each disk on the RAID will be requested by the same block over and over. This would ensure at some point those blocks do make it into the SSD.

Agreed. Different strokes for different folks.

 

In general yes. But for mobile desktops, it all depends. I'm not doing it w/ my x7200, as TRIM is not supported with SSDs in RAID using ICH10 and the Intel RST drivers, but I've been booting off a RAID 5 array on an older Clevo D900C for about 4 years now. Boot times are not bad 45-55 seconds on a 4 yr old laptop. The important thing was I never had to lose significant downtime on that laptop due to disk failure.

 

With the corollary that if one of your disks did fail, you'd be in much bigger trouble than if you were running a simple mirrored pair, or even a single hard drive & taking an image/backup daily.

(this is without considering the fact that array drives don't tend to die one at a time; this will be especially true for a portable computer where the disks are subject to the same external/environmental hazards & factors)

 

I'm ICH9 but did you try to be sure all power management was disabled? Such as even an outside app like Quiet Disk where apm and aam etc are set to 255? I there is a drive non conformance issue this would not help.

What I have found is some of the individual drives seem to vibrate where they under run rotational speed. They then will pause while they try and spin themselves back up. Again this is a best guess but what it looks like they do. If one of the three drives is doing this it would give a good reason why the system is getting degradation.

 

Good point on that all drives would be exposed to the same environmental stresses - although due to case design and location one of the drives may be in a safer location from "normal stresses" such as heat.

Not sure I understand your point regarding a single disk failure. In this case, shutting down, replacing the disk and rebuilding the array (in realtime) is all that is necessary. In the case of a single drive, downtime is greater in that you would not only replace the disk, but then wait while restoring the image.

 

No idea. But, that seems like one possibility. Thanks.

I'll bring it up in the other forum, as I definitely could see that affecting the participation of a drive in the RAID volume.

 

Yup, but that is predicated on always having a spare disk to hand, which is perhaps not applicable if we're talking about portable systems. The array will continue to function, albeit at a slower speed, but you're then in that very nerve-racking position of not knowing whether another disk will drop out before you can get your hands on a replacement. (it's for this reason that my storage server runs RAID6 w/ two hot spares per array, plus some cold spares in a drawer. Paranoid? I prefer 'prepared'.)

& yes a lot of the same applies re. imaging a single disk to say, an external hard drive, and the internal disk dying, but the crucial difference there is that your data is all in one image on the external drive, rather than split for parity & such by a RAID controller and therefore much more difficult to reconstruct if you get multiple drive failures.

 

Actually, due to the size of these drives, I have two drives for the old system - one one my shelf in the office, and one in the laptop's bag.

Agreed, thus the need for backups outside of the machine. Somewhere on the inter-webs, a post of mine exists stating, "RAID != BACKUP; RAID != BACKUP; RAID != BACKUP"

 

People certainly struggle with the difference between 'availability' & 'backup'...

Sadly my home array is used for backup to a certain extent - the really really important stuff goes to an online backup provider, & the fairly important stuff gets archived to hard drives or DVD-RW depending on size, but a huge amount of stuff just has to stay on the server because I can't afford to back it up anywhere else. Hence the two hot spares & running RAID6 instead of 5.

 

<aside>In the same boat on my home machine, until I plopped down $100 for an external 2TB drive. Only thing I'm bad about is taking the drive offsite in case of catastrophe.</aside>

 

I have >40TB of data, so doing it all on externals would be nightmarish...

 

40TB is a lot of pr0n

 

My media library informs me it will play for 785 years before it repeats itself.

Anyone know where to get good bulk (and I mean, really bulk) deals on toilet paper?

 

Well, how soft are we talking about here?

Do you just want cheap, or do you also want thick and/or plush toilet paper? Because the cheap stuff can be downright abusive if you use it every day.

 

A friend of mine once told me cheap toilet paper is a lot like <insert aging favorite tough guy actor here> - old, wrinkled, and won't take carp off of anyone.


Back onto the question at hand, a Seagate rep is looking at this on their end (see previous post in this thread). All I'm asking from them is a binary answer is RAID 5 supported with Momentus XTs? Hopefully they'll respond soon.

 

I finally found the page I knew I read many months ago:

XT's are not recommended for use in RAID (any version of it).

See:
Momentus XT APM spindown feature causes "lag spike... - Seagate Community Forums


Hope this helps.

 

That is what I've been looking for - something official from Seagate. I'll wait to see if Seagate comments on any of those later FW updates that may have fixed the problem.

Thanks again!

 

OK, so I am just getingin on this. I am the guy that had the issues, and it looks like the above post XT's are not raid friendly. I will try to answer some of the questions that came up....as I remember.

Yes, I am now running raid5 on WD Scorpio blacks, no issues so far.

As for firmware on the XT's , I am at SD24 on them, and FWIW they are now in a diff laptop performing just fine NON raid.

As to way Raid5 in a laptop, its for data integrity and uptime.......

As to the backups..I do them, nightly, I am anal about them....

Did I cover it all ?

 

TANWare,

wwoods has a solution, so I don't believe this is of importance to him right now.

I however, am running the XTs in a mirror. To this day, haven't had a problem (knock on silicon, mag disks, and NAND flash). However, I still think you are on to something, and tilleroftheearth's post from the Seagate forum seems to confirm.

I didn't install any tools to see if the Advanced Power Mgt could be turned off on the drives. However, I *did* go into Win 7's power mgt, and made sure the drives never power down - just in case that has any bearing.

It makes sense that if Intel RAID service queries for something from the drives, and they are not responding due to some power mgt setting, that the RAID controller would next report the drive as degraded.

 

Oh now just wait

This is something I am following Although, TBH I am pretty convinced its the XT's and raid5 that caused the problem, not vibration or power management....the XT's + Raid5 . I have replaced all of the with WD Scorpio Blacks and have had no issues at all.

Plus, I have been in contact daily with my reseller and on occassion Sager support, and while a XT + raid5 setup is not common, they have seen other issues like this with it.

Also, I have a email that I can not paste here, Due to forum rules, I am not alllowed to post personal correspondance, but at least one reseller is not going to be recommending or suggesting that someone use that configuration, and a follow up with Sager is being done to possibly remove that configuration option.

Is it possible that this effects only raid5 + the XT's, entirely possible, probable more likely because no one I have spoke to at the reseller level or Sager has seen this with anything BUT raid5. Is it possibly only this specific chipset, yup, sure is.

But I can say with almost 90% assurity in my case it was XT + Raid5, and my case, a few others and a link was enough to convince one reseller there is a issue with that combo.

 

Yes,

I agree it is something in the XTs, but what? What is causing the drive to go degraged? I wholeheartedly agree something about Intel RAID's query that the drive is not responding correctly and the driver think it is degraded. Also, searching around the net I found other posts with similar problems of this XT, but with RAID 0.

So we know something is there. But again, what? That is what I'm determined to figure out.

In that vein, it is possible the default power mgt settings of the XT is causing this problem. In tilleroftheearth's post, someone disabled APM (the default is on) on the drives, and it stopped giving them any problems in RAID. That is until they rebooted. After the reboot, they rechecked APM. Lo and behold, it was off again. The changes did not stick. I also noticed in the "power plan" in Win 7, the drives shut off after 20 minutes of inactivity, regardless of power supply. This would be the default on every computer.

So while there is definitely a problem with default configurations of XTs in RAID (5, and possibly 0), I'm crossing my fingers it is caused by a "configurable" option that a few tweaks or a FW update may control.

 

THAT, my friend is the 64,000$ question, and TBH I think only Seagate can answer that for 100% certanty......and TBH, I am not 100% sure Seagate knows at the moment.

 

See post here : http://forum.notebookreview.com/sag...ger-np7280-owners-lounge-419.html#post7291532

Near bottom, #4187, by XoticPC

 

The XT is known to do its own thing with APM, if you set it to 255 and AAM then it will not just shut down on its own normally. The problem also when it spins down is it will not give data all at once when it has to spin up. The nand is there to feed data, even partial data, untill the drive spins back up and provides the rest. If it can it will provide all the data from nand and leave the drive spun down.

This is part of the energy savings mode of the drive and of course will cause havoc in a raid array unless the drivers controling the stripe know of this. Since this is out of the norm for any other drive it is considered only partial data and the array is considered degraded.

So in the end the only way to stop raid issues is to prevent any type of spin down. Raid1 or possibly JBOD will not casue an issue but an array will. Now I had it, XT Raid0, for one day but also set APM, AAM to 255 whn I ran it.

 

I am not sure if one day would have been long enough to show issues, my issues dident crop up untill between 2-3 days

 

Agreed one day was short. I knew though because of the APM properties of the drives it could never spin down. Once a spin down occurs the NAND would stay live and leave me with an array going haywire. Also to prevent this the short term array was striped at 2KB per drive and formated to 4KB per sector forcing both drives active on all reads and writes. I killed it because of CPU over head at those low settings.

There is no way drivers let alone controlers are programed to handle the XT's live NAND and spun down drives in an array. where one drive is supplying a stream of data the other supplys lets say 1/2 the file stream then pauses to let the drive spin up and supply the rest of the data. The array software and most likely the controler itself will see this pause in the stream as drive degradation etc.........

Edit; in the end you can only array the XT's if you can garenty to never spin down. Since the XT's are not geared to run in this manner you will always be subject to an eventual failure of the array. If via FW or external drivers you can keep the platters spining, you are golden. Again though you still risk the problems of any array and since XT's long term failure rate is yet to be seen it becomes a large gamble..............

 

TANWare,

All good stuff, and it makes sense. In regards to the XTs in a RAID, you're saying the unknown MTBF of the drives that never shut down that make this a bit dangerous or at the very least unexplored territory? From what I've seen from Seagate, they're giving it a MTBF of 100000 hours.

Let me pose another question. I've noticed a small freeze in my browser which crops up every now and then within a day's work. I have all my local and system's TEMP directory on my XT mirror. After reading your first post, it got me thinking. What if the pause is related to the browser trying to get data off of the drive? So, I changed Win 7's power settings to NEVER shut down the drives when on AC power - I didn't mess w/ AAM. I'm on day 3 now w/ out any pauses.

I don't know what effect this will have on the the drive's lifespan, but since this laptop is only running 8-12 hours a day, I feel the drive should last well past the time it would be to upgrade.

Thoughts?

 

The biggest wear points for drives normally is spin up. Where the heads are not "Floating" yet. keeping them spun up should be ok as far as MTBF but of course bad for battery life. AAM usually relates to the heads and seek time.

The XT in the end is a laptop drive and meant to save battery as much as possible. In this way durring idle time it will spin the platters down but keep the NAND active. it does this to hopefully read needed info from nand and keep the platters spun down. Without repeating myself for an array this is horrible as data can be on one drive but not another according to the data size read and the stripe.

This can then have one drive spun up while the other isn't. Then a new data read gets data from say the nand of the two drives but then has to wait for one of the ndrives to spin up for the reast of the data while one is trying to supply the data. In your case the mirror drive shuts down then all of a sudden activates and spins up while your primary drive is already running.

So too properly use the XT in any kind of raid it should be treated like a server drive where it NEVER shuts down. Of course Seagate does not directly support this with the XT so I doubt they will support raid itself for the drive. Now they easily could release an Enteprise version for servers and raid at some point.

 

In the 5+ months I've had the x7200, I've been on battery maybe 3 times. So, there is definitely a difference in usage pattern for a Desktop Replacement laptop vs. a mobile laptop.

Actually, I think this is part of the problem even with a SINGLE disk, and is a likely explanation of all the ppl on the Seagate forums complaining about "stuttering", "freezes" or "lags."


In any case, this has been a good exchange, thanks for participating.