Announcement

Collapse
No announcement yet.

OCZ Agility SATA 2.0 SSD 120GB

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
    d2kx
    Senior Member

  • d2kx
    replied
    The Indilinx SSDs (OCZ, SuperTalent, etc.) are getting TRIM support within this week (SuperTalent has, OCZ gets tomorrow). Hopefully it'll be supported by Linux 2.6.32 then.

    Leave a comment:

  • Kano
    Kanotix Developer

  • Kano
    replied
    That's a good idea to bench a fresh and a used drive. Even if that TRIM command is not directly available in Linux i am sure that there are tools the can reset the drive state.

    Leave a comment:

  • paul_one
    Junior Member

  • paul_one
    replied
    That's because a 120G drive has perhaps 130G of blocks available to it.
    This extra space (I don't remember the exact amount of extra space given) allows the drives to write and move blocks about if it needs to combine many blocks to delete them etc, speeding it up.

    When dd-ing a drive a couple of times over, this area will probably get eaten into - while dd-ing a fresh drive only once will leave this spare space alone.

    You also have the firmware storing reference tables etc, and the more that get's messed up, the worse the performance is (another thing I guess 'GC' does on the newer firmwares).

    I've only JUST got 2 30G vertex drives right now (afraid to partition/format them as I'm not 100% sure if I have the 512K alignment correct) - and if these drives are as good as they look then I might get 2 of these and compare - although for a PC you'd want drives around the 120G mark.
    I could do some preliminary file creation / random file copy tests from a ramfs filesystem (so you get rid of that useless read / CPU bound problems) like I did to test my current Core2 (yes, I know - but they're RAID-ed and I'd prefer the nice new Vertex's in my laptops ta :P).

    It strikes me as odd that a Vertex couldn't out-perform a normal rotary drive in a few of these tests - it screams "something wrong here" to me.

    Anyway - PEACE!

    --
    Paul

    Leave a comment:

  • drees
    Junior Member

  • drees
    replied
    Originally posted by Jonno View Post
    Well, garbage collecting is great, but if all logical blocks are written to neither write combining nor garbage collection will be very usefull, as ther will be no room to manouver in. And without TRIM support all logical blocks will be written to in relatively short order (depending on disk usage of course). Combining this with TRIM support in a future firmware release will be great though. Perhaps even enough to match Intel X-25 M G1 performance, though only time will tell for sure.
    I don't know - it seems like the PCPer report shows that as long as the drives have enough time to GC, it is enough to regain full write performance. Is it less than optimal? Probably - I'd imagine that with TRIM support less time will be needed for GC.
    Originally posted by Jonno View Post
    Well, either that or both drives should have been dd:ed to the same "used" starting state. That would also make them comparable, though at lower (and more realisitic) performane.
    Even a dd wipe is not always enough to bring drives to the same state from differing starting states - I'm pretty sure that Anandtech and others found that out as well.

    Leave a comment:

  • Jonno
    Phoronix Member

  • Jonno
    replied
    Originally posted by drees View Post
    Also to note is that the 1.30 version of OCZ SSD firmware for the Agility and Vertex introduce a concept called "Garbage Collection". The GC routines kick in when the drive is otherwise idle and is used to organize/collect data on the flash so that it will be able to do it in the most efficient/fastest way possible. The 1.30 firmware has a very early version of GC that can take a lot of idle time to take effect. The next released firmware should be much better in this regard. PCPer has an article which tests an early release of this new firmware which shows how it can help.
    Well, garbage collecting is great, but if all logical blocks are written to neither write combining nor garbage collection will be very usefull, as ther will be no room to manouver in. And without TRIM support all logical blocks will be written to in relatively short order (depending on disk usage of course). Combining this with TRIM support in a future firmware release will be great though. Perhaps even enough to match Intel X-25 M G1 performance, though only time will tell for sure.

    Originally posted by drees View Post
    At the very least, both SSDs should have been wiped clean using the tools available from OCZ to make sure both were in the same starting state.
    Well, either that or both drives should have been dd:ed to the same "used" starting state. That would also make them comparable, though at lower (and more realisitic) performane.

    Leave a comment:

  • Jonno
    Phoronix Member

  • Jonno
    replied
    Originally posted by d2kx View Post
    TRIM is supported by ext4 (dunno about ext3) since Linux 2.6.30 or 2.6.31. But there is still no SSD avaible that supports this command. Firmware updates for Intel and probably the Indilinx are coming September/October-ish.
    While technically true, it's not the whole truth. While ext4 (but currently not ext3) does issue TRIM commands in what the developers believe is the correct places, the VFS layer in the kernel does not pass it down to the drive, as that is considered dangerous to do without testing it on actual hardware. So just updating the firmware when it's released won't be enough, you'll need a new kernel too. The VFS will defenitely not get TRIM support before 2.6.32, most likely not until 2.6.33. So most end users won't see it until the Spring 2010 release of their favorite distribution.

    Leave a comment:

  • drees
    Junior Member

  • drees
    replied
    Originally posted by Jonno View Post
    Firstly, the Agility series ships with the 1.30 firmware version, which is availible as an update to the Vertex, while your Vertex only was updated to 1.10.
    Yes.

    Originally posted by bugmenot View Post
    The reason why 120gb OCZ agility beats 60gb OCZ vertex is because they are different sizes.
    Yes.

    Thanks for pointing out these obvious flaws in this benchmark. Very, very sloppy.

    Also to note is that the 1.30 version of OCZ SSD firmware for the Agility and Vertex introduce a concept called "Garbage Collection". The GC routines kick in when the drive is otherwise idle and is used to organize/collect data on the flash so that it will be able to do it in the most efficient/fastest way possible. The 1.30 firmware has a very early version of GC that can take a lot of idle time to take effect. The next released firmware should be much better in this regard. PCPer has an article which tests an early release of this new firmware which shows how it can help.

    At the very least, both SSDs should have been wiped clean using the tools available from OCZ to make sure both were in the same starting state.

    Leave a comment:

  • Jonno
    Phoronix Member

  • Jonno
    replied
    Originally posted by tomm3h View Post
    Formatting does NOT reset the disk!
    Originally posted by Zhick View Post
    Then what would actually reset the disk?
    dd if=/dev/zero of=/dev/sdX ?
    No, as SSD disks differentiate between empty blocks and block with zeroes on it. To write data to an empty block is an easy operation, but to write data to a used block (including a zeroed block) requires the entire erase block (usually consisting of several hundred write blocks) containig it to be erased, and then rewritten again (with new data in that single changed write block). And after that operation all blocks are still used, so the whole procedure has to be done again for the next write block. Write caching will of course help somewhat, but only if multiple writes to the same erase block happens close enough to each other. This is the major factor making random writes hundreds of times slower than sequential writes on SSDs, even though there is no seek latency.

    TRIM is a relatively new addition to the ATA standard that tells the drive that you are not interested in a block any longer, and the drive can go ahead and erase it whenever it has nothing else to do (or just skip rewriting it the next time it is forced to rewrite the erase block). However, hardware vendors don't dare ship a firmware with TRIM support before they have tested that it works in Windows 7 (the first MS OS with TRIM support), and Linux developers don't dare add TRIM support to the kernel untill they have tested that it works on actuall drives. Becouse if you get TRIM wrong, you'll lose user data for sure, and no one likes that. Thus no TRIM support until later this fall.

    Future versions of mkfs.* will hopefully get TRIM support, so that reformating a drive realy does reset the drive to a brand new state, but right now it doesn't, it has no effect on the performance of a drive. Your dd trick, on the other hand, is actually a good way of forcing the drive to a "used" state where it will perfom at it's worst (becouse all blocks have been written to).

    As most disks won't stay brand new for long, "used" performance is actually way more insteresting than "new" performance, so Phoronix should have done that dd trick on both drives before conducting their tests (though I would have used /dev/random just in case the firmware tries to be smart with zeroed blocks, wich isn't likely, but not implausible either). After updating the Vertex firmware to 1.30 of course.

    As stated by others, disk sizes can have a speed impact as well, so comparin the same size would have been more honest, though comparing a 60GB Vertex with an only slightly more expensive 120GB Agility would be interesting as well, if all else was equal. As it stand, this particular benchmark is worse than useless however.

    Leave a comment:

  • d2kx
    Senior Member

  • d2kx
    replied
    Originally posted by Kano View Post
    New os like Win7 or maybe some Linux kernel versions allow the use of special delete commands to recover a used sector but i don't think that dd will use those...
    TRIM is supported by ext4 (dunno about ext3) since Linux 2.6.30 or 2.6.31. But there is still no SSD avaible that supports this command. Firmware updates for Intel and probably the Indilinx are coming September/October-ish.

    Leave a comment:

  • Kano
    Kanotix Developer

  • Kano
    replied
    As there are only a few real ssd manufacturers you can get the same with a different branding and price. Just a bit tricky to find out who created em. For intel ssds the intel logo will be still somewhere even if it was produced for oem, others may be harder to identify. OCZ was always oem - they bought ram modules and hided the real manufactors with a cooler. That's the usual way.

    Leave a comment:

Working...
X