Ordering an SSD, 2 questions (MTBF, Performance). Thanks!

vbgunz

Junior Member
Aug 27, 2010
16
0
66
I am in the market for an SSD. In particular, I am looking at the "OCZ Vertex 2 OCZSSD2-2VTXE60G 2.5" 60GB SATA II MLC Internal Solid State Drive (SSD)" from newegg here http://www.newegg.com/Product/Produc...82E16820227550

I am using Kubuntu 10.4. I wish to dedicate an SSD to the boot, root and swap on my system. My root in 5 years is at about 7GB total (not counting /tmp). I wish to dedicate 16GB of the SSD to swap. So I plan on immediately using 23GB as soon as I get it.

Here is my question. I hibernate up to 2 times a day and probably write up to 6GB a day doing this alone. The details on the disk at newegg state this "MTBF: 2,000,000 hours" and if I am correct, that's 228 years until failure. But consensus to wear, tear and failure seem quicker.

Q. How does it really work if on average I am writing say 10GB a day to the disk? My main concerns go into speedier boot, binary access and hibernate/resume. Am I gonna dog this disk?

One more question. When I was reading some articles on this site, I believe I saw the single most important factor was random writing to the disk. If this is correct and newegg details say this about the disk "4k Random Write (Aligned): 50,000 IOPS" is that good in terms of performance amongst SSDs?

Q. I just want to know if I am making a pretty good decision on this disk. I believe the price per GB is good here for me *but* is the performance what I think it is in comparison to other disk at newegg?

In other words, if you know of a better disk, I'd like to hear about it as I will most likely order this disk before the day is up. I assume, I can plug it right in like any other SATA HDD, boot, format to (Ext 4) and go (I hope).

Thank you for your time. I appreciate it!
 

sub.mesa

Senior member
Feb 16, 2010
611
0
0
16GB SWAP? You running some extremely memory consuming programs? This might work as the SSD is quite fast, but ordinarily you shouldn't need that much swap, if any at all.

Your /tmp should be tmpfs, in RAM memory. You shouldn't cause unnecessary writes on the NAND i think.

MTBF is fun, but assumes no firmware issues that brick the SSD and no power/physical issues; i.e. a perfect environment. I would look more at controller brand reputation:

Intel (best)
Indilinx (proven; stable)
Sandforce/Micron (had issues, may be good now)
Samsung: (not really sure where to put this one)
JMicron/Toshiba: bad

SSDs only wear if you write to them, so if it's going to be a mostly read-only device for the next 200 years then yes in theory it could reach that i think. But only in a perfect environment.

Some SSDs tend to fail instantly, without there being any apparent cause. Firmware issues or sensibility to power fluctuations may be the cause of this; SSDs are not yet matured like HDDs are.

If you want a Sandforce SSD, then pick one with 28% spare space, like 50GB/100GB models. Those physically have 64GiB/128GiB NAND, but only 50GB/100GB is visible. This may improve performance if you will be doing a lot of small writes, like when swapping.
 

FishAk

Senior member
Jun 13, 2010
987
0
0
MTBF is a useless marketing gimmick with no real world meaning or usefulness.

The Vertex 2, and in fact all SandForce based drives use a compression algorithm when writing data. This is one of the less mentioned advantages the SF controller has over all other controllers introduced to date. Obviously, this will positively effect the wear of the nand.
 
Last edited:

alaricljs

Golden Member
May 11, 2005
1,221
1
76
Oooh, sweet. A deal that finally beats the months running MWave $140 deal. I have this very same disk sitting in a Linux server running: Gentoo Xen 4, with 2 Gentoo guests and a WinXP guest.

What on earth do you need 16GB of swap for? Between the xen host and 2 guests I have a total of 3GB of swap. 1 guest is a fileserver (6TB atm), the other is a complete mail/web server with SQL/antispam/wiki/webmail/www(php)/calendar and more.

/tmp should definitely NOT be on disk. However it lands on disk if you need the RAM that is occupied by /tmp. It's really rather well designed. /tmp is directly attached to your RAM so you can do stuff like unzip/untar to /tmp, install or relocated the unzipped data and half of your IO is to RAM instead of hammering your disk. If you don't have enough RAM then /tmp is shunted over to swap.


And to round that all out, the drive is wickedly fast. Those random 4k write IOPS are exactly what you're looking for on compiles and SQL DBs.

As to what sub.mesa is saying about the higher spare reserve, I don't see any need for it. While I don't have one of those drives to test against I see no noticeable reduction in my drive's speed even when 90% full. Maybe it bench's a little slower, but I'm not running benchmarks I'm running a production server that sees no change in service times.
 
Last edited:

vbgunz

Junior Member
Aug 27, 2010
16
0
66
To address the 16GB of swap, well, maybe I am old school *but* I have 8GB of memory in my machine. The adage safe bet when it comes to swap is to simply double what you have in RAM. My aim at getting an SSD definitely involves 16GB being dedicated to it.

My usage is for the desktop/workstation. My biggest concern is hibernation and resuming from it without a problem. At night, the PC simply suspends to RAM. I expect to use at most 50% of the disk *but* am getting a 60GB over a 30GB or 40GB as there is no telling what's on the road ahead.

Cost wise, the 60GB vertex 2 seems to be the better deal Vs it's smaller counterparts. The 100GB vertex 2 in the bench here at anandtech seems to be the clear winner in performance. If this is the case, I am really leaning on getting this SSD within the next couple of hours.

I just wanted to know if my usage is gonna dog the disk and if performance wise, I am making a good decision on getting the vertex 2 in particular to anything else.

Thank you all for your help, I sincerely appreciate it!
 

Zap

Elite Member
Oct 13, 1999
22,377
2
81
If you're worried about longetivity, just keep backups and buy an SSD with a long warranty. Don't know if they extended it for other models, but the Patriot TorqX comes with a 10 year warranty.
 

alaricljs

Golden Member
May 11, 2005
1,221
1
76
Suspend to RAM and Hibernate are 2 totally different things. I have a Gentoo laptop that I S2R every night on the way home from work. It's been fine doing that for months at a time. I only need reboots to get into a new kernel.

Hibernate takes S2R's data out of the RAM and saves it to disk then totally powers off the system. Great for using long-term like if you went on vacation. You could even ship the system in that state.

Which are you doing? Personally for the time saved doing S2R and not hibernating is worth the possible extra cost of a few watts and the cost of the disk space even on a chunk of HDD. I just don't think hibernate is worth it over S2R. Then again I have a UPS, do you?

8GB of hibernation data to the disk every night is definitely going to put you on the upper end of usage. Take that 1 thing out and you're unlikely to wear out your drive before you want a bigger one or a fast enough one is available cheap enough for you to want to upgrade or cross-grade with a 2nd machine.
 

darckhart

Senior member
Jul 6, 2004
517
2
81
also, as a way-out-there choice (no one else seems to mention this since it's a drastic case), but i believe a secure erase will essentially reset your ssd's nand performance back to factory. and with a tiny 60gb drive, you might clone it somewhere temporarily and go this route when you get sick of things/performance 3-4 years down the line.
 

vbgunz

Junior Member
Aug 27, 2010
16
0
66
@alaricljs I hibernate up to twice a day as I dual boot into Windows 7 and I love returning to my previous state every time. I suspend to ram at night automatically and in between uses when my intent is to power down but start back up in Kubuntu and do it fast.

Hibernation/Resuming is probably my main selling point.
 

vbgunz

Junior Member
Aug 27, 2010
16
0
66
@alaricljs

I mostly play MP games e.g., L4D2. By the way, I never hibernate/suspend/resume Windows and don't expect for it to ever see the disk. This disk is targeted for use only in Kubuntu.
 

alaricljs

Golden Member
May 11, 2005
1,221
1
76
Fun! ;) I like having my Windows games box separate from the rest of my stuff.

The 8-16GB worth of writes for your hibernation plans is definitely going to put you outside the typical usage range if you're doing that consistently.

In 1 year, if you hibernate once on 292 days and a 2nd time 146 of those days and don't hibernate at all on 73 of those days (80%/40%/20%) then you will write 3.5TB of data that year just for hibernating. This equates to your currently planned empty space to be re-written (wear leveling) ~87 times.

Right now I can't find data on how much you're supposed to expect to be able to write to these things... use the word maximum in a search and all they want to tell you is about performance.
 

FishAk

Senior member
Jun 13, 2010
987
0
0
In 1 year, if you hibernate.....(80%/40%/20%) then you will write 3.5TB of data that year just for hibernating.

It would be very interesting to know if, since much of the same data will be written multiple times for hybernate, and with SF's unique algorithm for saving data, if the SF controller would actually rewrite those files each time, or if it would realize it's already got them, and just refer to them again. That would be cool.
 

alaricljs

Golden Member
May 11, 2005
1,221
1
76
It would be very interesting to know if, since much of the same data will be written multiple times for hybernate, and with SF's unique algorithm for saving data, if the SF controller would actually rewrite those files each time, or if it would realize it's already got them, and just refer to them again. That would be cool.


It's not files... It's the contents of your RAM. If you're usually running the same stuff then there's definitely going to be duplicate data but it won't be in the same location in the "file".

Would be very interesting to see the reality of what the SF controller does with that.
 

FishAk

Senior member
Jun 13, 2010
987
0
0
SSDs don't store the data like a HDD. They have no latency penalty, so they can put pages wherever they want to on the nand. I know you already know this, its part of the wear leveling strategy. It doesn't necessarily follow "files".

When Hibernate pulls the data off the RAM, it's always going to have lots of stuff that was the same as previous times- regardless of the exact behavior of the user during the session. If the user behaves similarly each time, obviously, more stuff will be the same.

I'm not sure where I read this, but the source said that installing Word and Excel to a SandForce drive, resulted in only 50 percent of the data being written, since so many of the files are so close to each other. If this is true, this is much more than simple compression. If the controller can remember these bits over several sessions, instead of just over an install, that would be huge.

Hmmm... maybe that's too many ifs.
 

alaricljs

Golden Member
May 11, 2005
1,221
1
76
Well, at work we use VTL (Virtual Tape Library) for our backup solution. It does block level de-duplication and the site I'm responsible for has the best de-dup statistics of all our sites at %40 stored data. This is impressive because the only thing we store is "data with business impact" that means it's basically all code and whatever binary gunk gets dumped out by ClearCase when you do a backup.

I would put money down that the marketdroids are being liberal when they say "Word and Excel"*, but even so block level de-dup is awesome and I'd love to find out that SF uses it. Allowing users to see and control to some extent this feature would make running same-OS VMs on SSDs a killer deal (imho).


* They probably mean buying retail Word and retail Excel separately and installing them as such. I'd put money down that there would be less duplication of data in Word/Excel installed from a complete Office package.
 

Voo

Golden Member
Feb 27, 2009
1,684
0
76
The 8-16GB worth of writes for your hibernation plans is definitely going to put you outside the typical usage range if you're doing that consistently.

In 1 year, if you hibernate once on 292 days and a 2nd time 146 of those days and don't hibernate at all on 73 of those days (80%/40%/20%) then you will write 3.5TB of data that year just for hibernating. This equates to your currently planned empty space to be re-written (wear leveling) ~87 times.
Intel quotes around 100gb/day for 5years and if we believe the floating numbers about write amplification and write cycles that's rather conservative.. that shouldn't be much of a problem
 

vbgunz

Junior Member
Aug 27, 2010
16
0
66
I went ahead and ordered the Vertex 2 from OCZ. I just want to put my / and /swap on there and be done with it. I honestly expect for 50% of this disk to go unused but decided on getting the 60GB because you never know.

I forgot what article it was on this site but an article none-the-less sold me on the idea of getting an SSD to improve performance. Well, heres hoping I can have a much snappier experience on the workstation. Fellas I thank you all for your input. It was all very much appreciated.

After a while of some usage, hopefully I can get back and write my own personal user experience or something and maybe enlighten a fellow forum dweller. Again, I wish to thank you all and anandtech.com for the insight and welcome. Major thanks fellas!
 

vbgunz

Junior Member
Aug 27, 2010
16
0
66
I got the OCZ Vertex 2 just 2 days ago and today all seems well. I'll do a comparison in the sense of my user experience so far. I have no benchmarks to prove any of this *but* I don't personally even care too much for real numbers. I only care about perception. So it doesn't matter if I can read at 200MB or 200GB so long as when I go to do something, it's done immediately.

Prior to the SSD I had a WDC 640GB 7200RPM disk as my boot and root. It took me 15 seconds to boot to the login. Once logged in I have a script that I manually call that populates my virtual desktops with a total of about 15 applications. That action takes about 20 seconds to complete. Hibernating takes about 1 minute to shutdown and resuming takes about 2 minutes to resume. Once resumed though, the desktop is the most sluggish thing imaginable for about another 2 minutes.

The SSD completely skips the loading screen when I boot. About 5 seconds to the login window. Once logged in, the manual script I call probably completes in the 5 to 10 second range. It's just quicker which is real nice. The biggest part for me is the hibernating. It takes about 30 seconds to hibernate. The even better part is resuming which takes less than a minute and once the desktop is up, it's really up and responsive as if I never hibernated at all. That rocks my socks off.

I'm pretty happy so far but at first setup of the disk, I was really pissed. I followed this http://www.ocztechnologyforum.com/f...Tips-tweaks-and-alignment&p=373226#post373226 and in the end I settled on a swap file Vs a swap partition. Hibernating though gave me nothing but problems. I did the whole swapon, fstab, kernel parameter for resume/swap_offset, etc and still just had problems.

I just went ahead and booted into the kubuntu 10.4 live CD and attempted to repartition the SSD with 2 partitions. 12GB towards swap and the rest dedicated to /. No matter how I used fdisk though, I kept getting the first partition to line up at cylinder 1 instead of 2. So I did the next best thing which worked. I created one large partition starting at cylinder 2 and then resized the partition in KDE partition manager. This worked. 44GB for / and 12GB for swap. The alignment seemed correct.

I installed Kubuntu right off the live CD into the 44GB partition and pointed to the swap partition. Booted up, restored some files and quickly began testing hibernation. Hibernation was working as flawlessly as it did before. It was much faster in comparison and I was just happy.

Here are some stats from my system so far. sda is the Vertex 2 and sdb is the WDC.

Code:
/dev/sda:
 Timing cached reads:   7256 MB in  2.00 seconds = 3628.58 MB/sec
 Timing buffered disk reads:  542 MB in  3.01 seconds = 180.11 MB/sec

/dev/sda1:
 Timing cached reads:   7634 MB in  2.00 seconds = 3818.50 MB/sec
 Timing buffered disk reads:  406 MB in  3.01 seconds = 135.10 MB/sec

/dev/sda2:
 Timing cached reads:   7526 MB in  2.00 seconds = 3763.82 MB/sec
 Timing buffered disk reads:  626 MB in  3.00 seconds = 208.54 MB/sec

/dev/sdb:
 Timing cached reads:   6946 MB in  2.00 seconds = 3473.57 MB/sec
 Timing buffered disk reads:  324 MB in  3.01 seconds = 107.59 MB/sec

/dev/sdb1:
 Timing cached reads:   7586 MB in  2.00 seconds = 3793.91 MB/sec
 Timing buffered disk reads:  324 MB in  3.00 seconds = 107.99 MB/sec                                                                                                                                                                                                                                                          
                                                                                                                                                                                                                                                                                                                               
/dev/sdb2:                                                                                                                                                                                                                                                                                                                     
 Timing cached reads:   7600 MB in  2.00 seconds = 3801.49 MB/sec                                                                                                                                                                                                                                                              
 Timing buffered disk reads:  226 MB in  3.00 seconds =  75.33 MB/sec

I have the wiper.sh script from OCZ as it helps with trimming *but* decided not to run it just yet. I am still running on the Ubuntu 2.6.32-24-generic kernel. I am using Ext4 with journaling and the only real tweak in fstab which has always been there is noatime. I have the latest SandForce firmware of 1.11 on the SSD.

I am sorry I couldn't be more technically detailed but I promised I would get back and I did. Am I happy. Yes. If you guys have tips, suggestions and the like, I'd like to know what you're thinking. I thank you fellas for your time again!
 

sub.mesa

Senior member
Feb 16, 2010
611
0
0
With kernel 2.6.33 you would have full TRIM support, so no need to run any scripts. Have you confirmed your partitions are aligned? That might be worthwhile. fdisk output should tell you the offset. Usually the data partition should start at 2048 sector offset or 1MiB; just like Vista and windows 7 do. But any multiple of 128KiB would be appropriate.
 

vbgunz

Junior Member
Aug 27, 2010
16
0
66
@sub.mesa

sda is the SSD and sdb is the HDD. This is what fdisk -l shows

Code:
Disk /dev/sdb: 640.1 GB, 640135028736 bytes
255 heads, 63 sectors/track, 77825 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk identifier: 0x000d5ffd

Device Boot      Start         End      Blocks   Id  System
/dev/sdb1               1       63465   509782581   83  Linux
/dev/sdb2   *       63466       77826   115347456    7  HPFS/NTFS

Disk /dev/sda: 60.0 GB, 60022480896 bytes
255 heads, 63 sectors/track, 7297 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes
Sector size (logical/physical): 512 bytes / 512 bytes                                                                                                          
I/O size (minimum/optimal): 512 bytes / 512 bytes                                                                                                              
Disk identifier: 0x0002ff11                                                                                                                                    
                                                                                                                                                               
   Device Boot      Start         End      Blocks   Id  System                                                                                                 
/dev/sda1   *           2        5732    46034257+  83  Linux                                                                                                  
/dev/sda2            5733        7297    12570862+  82  Linux swap / Solaris

Man, I am beginning to think I got the alignment wrong. Isn't cylinders supposed to be at 1024 or am I reading it wrong? If need be, I can reinstall as I do backups like a saint so I ain't worried about losing data *but* damn. If I am wrong, what tip can you offer for doing this right? One partition at 44GB and the next at 12GB?
 
Last edited:

alaricljs

Golden Member
May 11, 2005
1,221
1
76
Ya got it wrong...

Code:
Disk /dev/sda: 60.0 GB, 60022480896 bytes
255 heads, 63 sectors/track, 7297 cylinders, total 117231408 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk identifier: 0x4711d693

   Device Boot      Start         End      Blocks   Id  System
/dev/sda1            2048      204799      101376   83  Linux
/dev/sda2          204800    16982015     8388608   fd  Linux raid autodetect
/dev/sda3        16982016    21176319     2097152   fd  Linux raid autodetect
/dev/sda4        21176320   117231407    48027544   fd  Linux raid autodetect

What I find most interesting is your "Units" line... wtf ?
 
Last edited:

dmoney1980

Platinum Member
Jan 17, 2008
2,471
38
91
Intel quotes around 100gb/day for 5years and if we believe the floating numbers about write amplification and write cycles that's rather conservative.. that shouldn't be much of a problem

quick question- so if your using an SSD, you should have disable hibernate (which everybody says you should do), but can I keep sleep on ? If so, what type of sleep should I use when in the BIOS ? (STR3?)
 

alaricljs

Golden Member
May 11, 2005
1,221
1
76
Ok, so are you using:
Code:
fdisk -l -u=cylinders
explicitly? If you are then don't, if you're not then you might want to find out wtf is setting it to use cylinders instead of sectors.

You want to use sectors which is normally the default for fdisk. That will give you proper output from fdisk to know if you're aligned. If you're not aligned it'll also allow you to do it manually.