Question for CAT MAKER - Concerning LargeSystemCache

[GamblerFEXonlin]

thanks for some reports Quitch and 99 to life.

With some of the workarounds in place my system was stable for a while but failed.

SO NV cards also can fail?? damn. It would be nice with more reports, I'd really love large system caache enabled... unzipping and loading maps between games inst the same anymore.

[Drumphil]

yep, if you are annoyed by this, petition microsoft to fix it.

[Sivar]

Greetings. I am the person who wrote the FAQ in question.
May I recommend that you view the StorageReview thread regarding this post here (http://forums.storagereview.net/index.php?showtopic=15009).

Wilst I found this:
http://faq.storagereview.com/tiki-index.php?page=SingleDriveVsRaid0
I still did not see where this guy had much proof at all to support his "hypothesis" especially concerning 2 threads hitting the array at once. Even with a single drive setup, 1 for OS, 1 for applications, you would still see that if you ran 2 applications at once the head would still have to hit 2 places on the same drive in the order that they requested, which doesn't give you anything except common IDE performance.
Loading multiple applications at once is largely seek-limited. Improved STR (Sustained Transfer Rate), which is more or less the only performance benefit that RAID0 provides, would be of very little benefit. May I recommend that you read StorageReview's Reference Guide (http://storagereview.com/guide/guide_index.html), to learn about the factors that influence drive performance? It is rather long, so you may want to read only the "Hard Disk General Performance Issues" portion.
The FAQ has some useful entries as well, particularly the well written Performance Factors (http://faq.storagereview.com/tiki-index.php?page=PerformanceFactors) by Eugene Ra.

And SCSI is frankly not recommended in my opinion for a board that does not contain a PCI66 slot. The PCI bus total can only handle up to 133MB/s, and having a U160 or U320 on there would saturate the bus, leaving Windows crawling. I will address this in my next reply.

I was not impressed by that article at all, the only part that is useful is concerning the pagefile but I already knew this.
You are more than welcome to suggest changes to any FAQ entry.

[Sivar]

And SCSI is frankly not recommended in my opinion for a board that does not contain a PCI66 slot. The PCI bus total can only handle up to 133MB/s, and having a U160 or U320 on there would saturate the bus, leaving Windows crawling.
I could see how that would be true if a person was using a PCI33 card for their RAID, considering the limit for PCI's 33 spec is 133mhz aka 133MB/s, and you would never be able to reach that IRL.
From reading several posts, it appears that many people are under the impression that the peak bus speed of a hard drive's interface (such as 133MB/sec for ATA133, 320MB/sec for U320 SCSI) is at all a significant factor in desktop hard drive performance, and that therefore PCI's bandwidth limitation is a major problem. This is simply not the case.

Even under optimal conditions, modern hard drives are not capable of even approaching their peak bus speeds. The fastest hard drive in the world (which isn't under NDA), the Fujitsu MAS3735, peaks (http://storagereview.com/articles/200304/20030429MAS3735_2.html) at just under 80MB/sec on the first half of its capacity. Even under the rare conditions where the drive can achieve peak STR, this is less than 1/4 of the bus bandwidth of U320SCSI, and about 2/3 of the real-world bandwidth of many 32-bit 33MHz PCI implementations. Even ATA100 would have no problem handling this, and ATA66 would be able to handle this drive's performance without problem for most non-STR-limited tasks.

Other than for large linear access (such as editing video), the transfer rates that a drive can achieve in practice will be lower still, since seek times and rotational latency have a huge impact on performance (even on a 15,000RPM enterprise drive). Even the simplest actions, such as loading an office program or game, or compiling source code, can send the actuators flying all over the disk. It largely does not matter what a drive's peak sustained transfer rate is, if it is only reading a megabyte or two (or less) before the actuator has to move again and wait for the platter to rotate the needed data under the heads. This is also one reason why defragmentation is crucial on filesystems which are not designed to do it themselves. (It is generally not needed outside of Microsoft Windows systems).

RAID (of any level) does not improve seek times, so even a RAID 1+0 array of four fast drives will not be limited by PCI bandwidth for most tasks.

That said, PCI bandwidth can be a limiting factor when a system is doing multiple heavy I/O tasks. For example, if you were using a PCI TV recording card, downloading files from your home server over gigabit ethernet to hard drive, and loading a program from your hard drive, then the PCI bandwidth limit would almost certainly be a major bottleneck. Just gigabit ethernet can be a huge bandwidth hog.
Another example would be a video editing workstation, since some types of editing are linear enough to allow a hard drive to "spread its STR wings."
If someone had even a two-drive array of, say, Western Digital Raptors, and wrote video to disk, they alone may not saturate the PCI bus but they would at the very least leave little bandwidth for anything else.

I use onboard SATA RAID right now, which does not suffer near the amount of speed issues that a PCI33 card would have.Use of an onboard drive controller is only beneficial when it is integrated into your motherboard's chipset. My motherboard, for example, has an integrated SATA controller, but it is a seperate Silicon Image chip. Usually, these seperate chips will be connected to the PCI bus just like a card.
Many new chipsets (which you may have) do genuinely integrate SATA, and almost all of them integrate ordinary PATA. In these cases, the only limitation of bandwidth is the link between the chipset's southbridge and its northbridge (if they are separate chips), and there are almost never enough ports to allow for even the possibility of this being a problem.

One extra note: Having a hard drive bus that is faster than the hard drive can achieve can still be of some small benefit. While the drive itself may be incapable of such speeds, the drive's buffers are DRAM, and can burst small amounts of data very quickly. In practice, the difference is at best a few percent, but it does exist.