mejv

Member
  • Content Count

    259
  • Joined

  • Last visited

Everything posted by mejv

  1. mejv

    3ware and oce and windows 2003sp1

    It is to be expected: The unit go expanded to the whole capacity, but the windows partition you had was set to whatever capacity you had originally... The expansion is not expanding the Operating Logical Partition... Use Partition Magic or the like to expand you logical partition... Good luck! MEJV
  2. mejv

    Dead software raid

    If it is a soft-raid... (What software are you talking about and what OS?) it should be not hardware related... so it should work on another motherboard with the same software and config... MEJV
  3. Was it writing data over the Gb network to the controller? If it is, this is another prof that the Atto benchmark is not a good one... Otherwise, it remains to be tested! MEJV
  4. Have you tried other benchmark programs? How about a real-life transfer timed ? MEJV
  5. Q11690 - Performance/Optimization: I am using the Atto Bench32.exe benchmark under Windows to measure performance with my 9500S, 9550SX, or 9590SE controllers. Is there any way to increase the Atto performance numbers? Note: The below description includes all supported 32-bit versions of Windows (2000/XP/2003). Information for the 9550SX below also applies to the 9550SXU and 9590SE. WARNING: The buffering registry keys should not be used with 64-bit versions of Windows. Doing so could cause loss of data and/or a system crash. There are two things you can change on the 9500S and 9550SX to tune performance under Windows: -Cache control setting for FUA (Force Unit Access) FUA commands are a way that the RAID controller or a program (such as a database program) can ensure that data is actually written to the disk drive media, and is not stored in cache. When a write command is followed with a FUA command, then the disk drive will only issue “command complete†to the controller once the data is written to media. When performance is considered more important than protection, it may be desirable to ignore FUA commands. By default the 9500S honors FUA commands. Using the registry key 9500S_cache_control_no_FUA_cache_setting_(best_write_performance).reg will cause the 9500S to ignore FUA commands from the operating system. Benchmarks such as Atto Bench32.exe that send the FUA command will show higher write performance numbers when using this registry entry. Using the registry key of 9500S_cache_control_(default_setting).reg will set the 9500S back to its default of honoring FUA commands. The 9550SX uses 'profiles' to set the behavior of the controller to FUA commands. You should use the 9550SX profiles to configure the FUA behavior, and not registry key. The Protection and Balanced profiles of the 9550SX honor FUA commands; the Performance profile ignores them. If you use a battery backup unit (BBU), FUA is ignored, because the BBU preserves the contents of the controller cache memory for a limited period of time (up to 72 hours), in the event of a power failure. Consult the 9550SX User's Guide http://www.3ware.com/support/userdocs.asp for more information. -Buffering size By default the 9500S uses a buffer for reads and writes of 128K. By default the 9550SX uses a buffer size of 512K, and a buffer size of 256K for Windows XP when booting from the 9550SX. Using the registry key of 95xx_buffer_size_4096K_(faster_readwrite_performance).reg will increase the buffer size to 4096K. Note that changing the buffer size to 4096K may affect the ability of the system to hibernate. Use the registry key of 9500S_buffer_size_128K_(default_setting).reg to put the 9500S back to the default buffer size. Use the key 9550SX_buffer_size_512K_(default_setting).reg unless using Windows XP and booting from the 9550SX. Use the key 9550SX_buffer_size_256K_(default_setting_for_WinXP_when_booting_from_controller).reg if using Windows XP and booting from the 9550SX. The registry keys to make these changes are attached to this KB article. Double click on the registry key to install it. A reboot is required for the changes to take effect. Article Attachments 95xx_buffer_size_4096K_(faster_readwrite_performance).reg 9550SX_buffer_size_512K_(default_setting).reg 9500S_buffer_size_128K_(default_setting).reg 9500S_cache_control_(default_setting).reg 9500S_cache_control_no_FUA_cache_setting_(best_write_performance).reg 9550SX_buffer_size_256K_(default_setting_for_WinXP_when_booting_from_controller).reg
  6. mejv

    Anyone with a 3Ware card?

    Are you sure you have the latest MotherBoard BIOS and latest 3ware firmware? Regards, MEJV
  7. mejv

    Need help with 3ware 9500s in RAID5

    Have you tested the raw speed of the 3ware array? Do you have the cache of the Device Manager->Disk Drives->AMCCxxx->properties In the Policies options check the Enable Advanced Performance and see if it helps... Good luck! MEJV
  8. Have you tried going into the 3ware disk properties and enabling the performance caching ? This prevent the OS from sending the FUA bit (Force Unit Access) to the disk (bit ignored by ATA drives but 3ware has a SCSI interface to the OS). As it says, the FUA force the data to be written to the disk before completing the command and executing another one... This delais the next commands and make the disk loose a revolution for every sequential write... Good luck! MEJV By 3ware disk properties, I meant Windows disk properties for the 3ware disk ... MEJV
  9. mejv

    ARC-12** vs 3ware 9590SE-*ML

    I'd go for the least expensive... Both have good perf. but 3ware has been in business longer, stable drivers... Good luck! MEJV
  10. Maybe 3ware does not want to implement SMART because it was never standardized, and that the vendor specific SMART stuff is not easily made available! MEJV
  11. It means that the drive cannot come ready, so it is not ready to read/write data over the connected cable... Check if the drive is spinning, connectors plugged in properly... Maybe download a utility from the manufacturor to check the health of the drive... Good luck, MEJV
  12. mejv

    3Gb/s SATA?

    Oh bfg9000 - You can't use your speedometer to it's full extent? Mayby you should upgrade the engine! Cheers! Hey! Isn't the speedometer in km/h... It still seem to be bellow 200mph! MEJV
  13. The 3ware 9550SX only works on 3.3v, not 5v... MEJV
  14. 3ware controller did not support any *cq until recently with the 9550SX... Are the controller updated to the latest firmware ? MEJV
  15. Hello there! I hope I can make this clear for you: What make the SCSI advantage in multi-user is hte command queuing. The SATA Raptor 36 and 74 GB use the ATA Tag Command Queuing that did not give much advantage due to the design. The big advantage of the ATA interface is the simplicity and low command overhead, along with good cache algorithm tuned for single command/ single user/desktop environment. The SCSI interface has been around for a long time with Command Queuing (up to 256 commands) tuned for maximum IO/sec. With a high amount of command queued, the drive can maximize the execution of the commands by re-ordering them according to the destinations and rotational location of the data. Imagine having a list of 256 destinations to go to with your car... You'd find a way to go to the closest one first instead of blindly going from one place to the other... same basic principal... Hope that helps! MEJV
  16. mejv

    Secure erase and G-list

    The SCSI Format command has an option that can be set to use the P (Factory) and G (grown) list of defect. The P list cannot be deleted by the command, only the G list can be ignored... But you need to be able to send this command with the specific bits... for exact cmd, look at : www.t10.org Draft Standards and Technical Reports Command Sets one of the SCSI-3 Blocks Command (SBC) documents... (extract of the sbc2 standard document:) 5.2.1 FORMAT UNIT command overview The FORMAT UNIT command (see table 11) requests that the device server format the medium into application client accessible logical blocks as specified in the number of blocks and block length values received in the last mode parameter block descriptor (see 6.3.2) in a MODE SELECT command (see SPC-3). In addition, the device server may certify the medium and create control structures for the management of the medium and defects. The degree that the medium is altered by this command is vendor-specific. . . . The simplest form of the FORMAT UNIT command (i.e., a FORMAT UNIT command with no parameter data) accomplishes medium formatting with little application client control over defect management. The device server implementation determines the degree of defect management that is to be performed. Additional forms of this command increase the application client's control over defect management. The application client may specify: a) defect list(s) to be used; defect locations; c) that logical unit certification be enabled; and d) exception handling in the event that defect lists are not accessible. . . . A format data (FMTDATA) bit set to zero specifies that no parameter list be transferred from the data-out buffer. A FMTDATA bit set to one specifies that the FORMAT UNIT parameter list (see table 13) shall be transferred from the data-out buffer. The parameter list consists of a parameter list header, followed by an optional initialization pattern descriptor, followed by an optional defect list. A complete list (CMPLST) bit set to zero specifies that the defect list included in the FORMAT UNIT parameter list shall be used in an addition to the existing list of defects. As a result, the device server shall construct a new GLIST (see 4.8) that contains: a) the existing GLIST; the DLIST, if it is sent by the application client; and c) the CLIST, if certification is enabled (i.e., the device server may add any defects it detects during the format operation). A CMPLST bit set to one specifies that the defect list included in the FORMAT UNIT parameter list is a complete list of defects. Any existing defect list except the PLIST shall be ignored by the device server. As a result, the device server shall construct a new GLIST (see 4.8) that contains: a) the DLIST, if it is sent by the application client; and the CLIST, if certification is enabled (i.e., the device server may add any defects it detects during the format operation). If the FMTDATA bit is set to zero, the CMPLIST bit shall be ignored. The DEFECT LIST FORMAT field specifies the format of the address descriptors in the defect list if the FMTDATA bit is set to one (see table 12).... Good luck! MEJV
  17. mejv

    Need advice on Raid-1 Rebuild times

    If you think about it, it should be a matter of reading the 'good' drives, generating either the DATA or the PARITY stripe on the rebuilding drive... so the speed should be somewhat less than the claimed sustained RAID speed (for RAID 5), but for RAID 1, one would think the drive average write speed should be the getting factor... MEJV
  18. mejv

    Slow external drive response

    Maybe the drive fell asleep..., took about 10 seconds to spin-up ? MEJV
  19. mejv

    Three questions...

    It does, however, 8,589,934,592 Tera Bytes is the limit (2^64 sectors with 512 bytes/sector). It will take a while to max it out... but who knows? maybe in a few years! MEJV
  20. mejv

    Three questions...

    For 2) ATA/SATA did increase the LBA addressing to 48 bits (not to worry too much, since it would max out at 256 TB on a drive... So far, the biggest HDDs are 0.5 TB (500 GB)... LBA 64 bits are a 'limitation' of SCSI. The 10 byte commands are limited to 32 bit (2 TB max) which is a limitation on the 32 bit OSs from Micro$oft. Only win 2003 SP1 and 64 bit versions with a service pack address that issue. (linux kernel 2.4 also is limited to 32 bit LBA addressing, kernel 2.6 fixed that...) MEJV
  21. I believe the N of the controller is for Narrow SCSI or 8 bit wide. This would reduce the speed to half of the speed possible, hence 80 MB/s. The speed seen of 40 MB/s seems to be a hard set speed of ultra2 narrow (80MB/s /2 = 40) in LVD mode It could be due to a speed negociation at power-up, not able to max out for some raisons already mentionned : cable or terminators (double make sure the drive does not have a jumper set somewhere...) Good luck, MEJV
  22. Well, 3 Gb/s is the amont of bits transfered over the wire in one second. There is a 8bit/10bit encoding scheme that occured, so each byte become a 10 bit value transfered over the wire... which make the 3Gb/s become 300 MB/s. as far as I recollect, a Data frame can be up to 8kB in SATA and 1 KBin SAS, separated with some words (word = 4 bytes) which effectively adds to the overhead lowering the transfer rate over a 1 second average... ...but like parallele SCSI at 320 MB/s, there is an overhead as well... 300 MB/s is therefore as relevent as 320 MB/s is for // SCSI, or 133MB/s for PATA... Conclusion, it depends wether you're talking about the 'burst data xfer speed' versus cable data the tranfer rate. As far as I know, nobody transfered 80 MB/s with ultra2 (SCSI LVD) on an ultra 2 controller and a 15kRPM latest generation SCSI drive capable of ~100MB/s sustained xfer speed... MEJV
  23. mejv

    Slow write speeds on 36GB MAU

    F.U.A.: Force Unit Accept, maybe! unless you can look at the CDB being sent to the drive... But SCSI drive honor this, ATA do not have such concept. Faster, but in case of power failure, no garanty the data was actually written to the media! MEJV
  24. mejv

    New 200GB Maxtor only sees 32GB

    Have you checked the jumper settings on the drive ? Some older controller bios need an update too... MEJV
  25. http://www.3ware.com/about/news/9550SXLaun...ressRelease.pdf Anyone got any test on this one ? pci-x 133 (finally) is 3ware catching up ? Claims of 800 MB/s for reads, 380 MB/s writes on Raid 5... verified ? What's your take on that one ? MEJV