Brad.B

Member
  • Content Count

    79
  • Joined

  • Last visited

Community Reputation

0 Neutral

About Brad.B

  • Rank
    Member
  1. Given that I'm getting equally sub-par (in my opinion) results from the motherboard's on-board controllers - the Marvell and the ICH7R - I think I'm going to call Intel instead, no? Here's HD Tach, RAID-0 (x4) on the onboard ICH7R, variable zone test: Here's HD Tach, RAID-0 (x4) on the onboard Marvell, variable zone test: FWIW the 8mb and 32mb tests more or less mimic the 3ware trends, except for the Marvell which showed dismal write performance. Here's the Marvell, RAID-0 (x4), 8mb and 32mb zone tests in that order: -Brad
  2. I'm actually experiencing similar issues, though with some variations. I put together RAID 10 and RAID 0 (across four WD1500ADFD drives) configs on a 4-port 3ware 9650 SE as well as the onboard ICH7R and Marvell RAID controllers built into my D975XBX2. When doing the very long-running variable zone test I get flat lines that are maxed around 115MB/s no matter which controller. The 8MB and 32MB tests on the other hand were showing wildy varying transfer rates across the container span, reaching 165 and higher but plummeting to 80 or 100MB/s in several places. I don't really understand this, and I'm ready to call Intel and ask them why I can't pull more than 190 out of a 4-disk stripe set that should theoretically max out at nearly 360. -Brad
  3. It gets more disappointing... I've tried a 4-drive RAID-0 on the on-board Marvell and ICH7R as well as the 3ware RAID controllers, and all seem to be hitting their heads on similar bandwidth ceilings, except for the Marvell which was showing especially dismal write numbers. I can squeeze reads as high as 192 MB/s out of HD_Speed if I set the block size to 1MB or more. -Brad
  4. Here's the third HD Tach run as an addendum to my last reply to you, HD Tach's variable setting on ICH7R RAID 0 128Kb stripe, all default settings. Yet another perplexing result in my opinion, being that it's pegged at 120 MB/s pretty much all the way across instead the other two runs which ran between 120 and 155 or so. As a side note, CPU utilization on the ICH7R is not bad at all! -Brad
  5. Hehe, borked indeed. No file system, all Windows and RAID options at default, i.e. Windows write cache on, RAID cache on, NCQ on, 64K. The pics are HD Tach 8mb, 32mb, then variable, then HD Tune. Results from the ICH7R RAID 0 with two drives and 128K stripe are somewhat intriguing themselves, though not nearly as weird as with the 3ware. Here's a set of results from HD Tach, 8mb, 32mb settings. I have a variable run going right now and can post that when it's done. Oh and FYI, getting back to the 3ware, I did try the card both in the PCIe-X4 and -X8/16 slot and there was no difference.
  6. I'm experimenting with a set of four WD1500ADFD Raptors, the on-board ICH7R RAID controller on an Intel Bad Axe 2 motherboard, and a 3ware 9650SE 4-port RAID controller with BBU. I'm seeing bizarre results in HD Tach RW and wonder if I should toss it and find something more useful. (WinXP SP2, FWIW). In JBOD each controller shows the drives with read speeds starting at 90 MB/s and trailing off to just under 60. Fine. If I turn off Windows' write caching, the write speeds are stuck at 10 all the way across (!). With Windows' write caching left on, write speeds start at 55 and trail off to around 40 in the 8MB test, and fare somewhat closer to the read speeds in the 32MB test and perform almost identical to the read speeds in the variable zone test. Results on a 3ware RAID 10 set are more bizarre. The screen shots below are read only, with the variable zone test showing what looks like a hard ceiling across the container at 110 for reads even though burst speeds are reaching 230, and read speeds that vary wildly from 85 to over 160 in the 8MB and 32MB tests. HD_Speed showed a more sensible 140-165 read rate depending on zone. HD Tune's results shown here only confused matters more. I'm only scratching the surface but before I go on with dozens of permutations I really need a sanity check on how best to determine the storage subsystem's raw performance. Ultimately I'm trying to determine how much difference there is between the ICH7R and the 9650/BBU, in RAID 0, 1 and 10, determine if either controller is susceptible to bottlenecks of any kind and get a grip on the relative value of the 9650/BBU. And what do I do with the various settings during testing? The HD properties in Windows show settings for write caching and NCQ. The 3ware has its own cache of course but also its own NCQ setting! What's combos would be best for real-life performance vs benchmarking? -Brad
  7. Brad.B

    Good alternative to front page

    Microsoft still supports Frontpage 2003 (although it became a separate product instead of being bundled with the Office "pro" packages) and if you wish to stick with Microsoft tools, Microsoft sells Expression Web. An upgrade version is only $90 at Amazon. The non-upgrade version will cost three times as much. -Brad
  8. The fastest SATA hard drive on the market is WD's Raptor 150, which is not SATA-II. Don't be a marketing tool. If you want faster you'll have to resort to 15K SCSI and even then, the performance benefit for your usage pattern may not be clear. The Raptors are fast (see first answer). Any drives in RAID-0 will be faster than they are standalone, at the risk of (exponentially?) increased probability of a single failure disabling the entire volume. An good off-motherboard RAID controller should help performance slightly by offloading CPU cycles, and perhaps even more by the ability to host its own cache - up to 1GB on some controllers. Whether it's really necessary is questionable however. Adobe is pretty good about prefetching for good perceived performance. -Brad
  9. It seems like the last few floppy drives I've had, have been pretty much disposable items. The last a little while and that's it. Since they're not used often I do always blow the dust out first, so that's not the issue. What are the most trustworthy brands and/or models? Thanks, -Brad
  10. Brad.B

    anyone well versed on Symantec AV Corp?

    Wow, you haven't checked the memory footprint, have ya'. If you don't think that behavior is correct then a re-install is in order. Which oughta' be easy with a CE rollout. Although if realtime scanning is enabled, which it always should be, all you have to do is click on a file in Explorer and SAV checks it anyway. -Brad
  11. Brad.B

    What to do with old hard drives

    Try this: http://ohlssonvox.8k.com/fdd_raid.htm
  12. Brad.B

    Dropped HD on floor. How to fix

    I'm re-reading the thread now and I don't see what you're talking about. Only ONE person - jboles - suggested professional services. I missed that post on the first scan. Otherwise there are posts from you suggesting the jumpers and cabling, describing sounds of failure modes, describing shock forces and risks, ET mentioning more amusing ways of destroying data (does suggesting transporting the HD in dirty socks really constitute a recommendation or even an implication?), and ultimately you telling the guy not to open the cover "but if you do"... Nice. I make no assumptions about the value or the nature of the data. Value is subjective. Potential costs are fairly up-front and I leave the task of weighing the value, up to the victim. Value is subjective. Things are worth what people are willing to pay. My brother recently hesitated to pay the US$800 or so that it would have cost to get back four or five years of family pictures. If I were one his kids I'd be scarred for life knowing that he even had to think twice about it. The big thing in my opinion is to mention the option of professional recovery as quickly as possible because hard drive failures, specifically ones which involve platter/head collisions, can snowball (figuratively and literally) and the longer a drive is operated the less change of recovery there is. Are squirrels and pirates a Finnish thing or is it just you? And what's the URL? ;-) -Brad
  13. First of all, get thee to Canon's web site where you can find all the drivers for Windows and OS-X. Secondly, if you want to share a USB printer on your LAN my inclination would be to find out if an H-P JetDirect en3700 will function with your printer, and do all the printing via TCP/IP. -Brad
  14. Brad.B

    Dropped HD on floor. How to fix

    I have now opened it, and there is nothing to notice. It looks perfect... From now on I will be a lot more careful with hard drives =) Aw, geez, I don't pay attention to the forum for a couple of days and it goes to heck. Am I crazy or did NOBODY here suggest using a professional recovery facility, the likes of OnTrack, etc. BEFORE the poor guy opened up his hard drive and pretty much dashed all hope? -Brad
  15. Brad.B

    Slow writing on MAP3367 RAID-0

    Sorry, when you didn't say which A8V exactly, I had to assume it was something more current. That you say you had problems low-level formatting the RAID makes me nervous right away. Just to be sure we're on the same level... The low-level format in SCSI is done from the controller, often during post but occasionally from within the software interface for the SCSI controller. The HBA abstracts all that stuff from Windows when you have a SCSI system so Windows only knows a file system format. None of that may matter however... PCI has a mathematical bandwidth limit of roughly 132 MB/s. In actual use PCI is frequently said to choke at numbers over 80 MB/s so for WinBench to be able to achieve 105 MB/s over PCI is actually remarkable. The bigger problem is that PCI bandwidth (which I'll reiterate, is not full duplex) is also shared across all PCI slots, so when copying data to and from PCI devices on the same bus, you're then subject to at best, halving your results. Remember too that the PCI bus isn't just the slots on the motherboard either. Factor in the overhead for ack/nack type stuff and I think you may be lucky to be getting the numbers you have! And yes, I know, you said the IDE HDs are connected to the Southbridge controllers, not the Promise. But until someone shows me a block diagram that proves otherwise I'm still blaming PCI bottlenecks and arbitration times. -Brad