Re: Experience with 18G IDE HD
On Fri, Dec 17, 1999 at 01:51:54AM +0200, Evgeny Stambulchik wrote: On 15-Dec-99 Eli Marmor wrote: As Gavrie noted, there is no difference between the technologies which should cause any difference in quality. The mechanics is similar. As far as performance is considered, it's not RPM that usually matters (inspite of a big hype), but the amount of cache on disk. Until recently (1-2 years), the maximum cache on IDE disks was 128KB, while about each SCSI disk had 0.5MB. This is not exactly the truth ... Check out reviews and some real numbers about Quanttum Fireball SE . It was UDMA/33 IDE drive whihc had only 128K cache (not a long time ago - two years) and it performed better than its WD or even IBM counterparts , which had 4-times bigger cache - 512K . I'm not going to say that cahce isn't playing any role in all that, it actually does, but when we're speaking about the real performance, we're speaking about continuous transfers of VERY big files - i.e. we're measuring the real read/write speed of drive, when it physically accesses the plates and transfers tthe data from/to it . The cache is important for relatively small files , which happen to be PHYSICALLY close to each other . For example, the cache controller of one of Qunatum drives (I think it was Bigfoot) makes caching based on very simple criteria - it analyzes which physical sectors of disk are mostly used for some period of time and caches data from these sectors any time it can, but be honest, nowadays even 2MB cache isn't enough to cache all files being used, not to mention that *real* OS sometimes swaps data, it's multiuser and multitasking system and almost at any time there is no cache hit :( . Just to show you how small 2MB cache is, think about this - minimal amount of data which can be read/written by HDD controller is 512bytes (size of one sector) and how many such sectors we have in 2MB ? 2048, isn't it ? So, how much sectors you can cache (you - hdd cache controller) in vast majority of cases depends on amount of small files frequently used by system (ideal variant) OR how many files are happen to be close enough each to other on hard disk . MOst of standalone test programs show that in case you're using somethine like 20-30 small files (less than 4-8K in size) you will have a cache hit , but if you're using "just" 20% big files (more than 100K approx) you will see that number of cache hits decreases by 40-50% . So, for real work , or for SERVER or for HIGH-PERFORMANCE workstation what is more important is the RPM , which means HIGHER linear speed no matter size of files being used most of time . Once I was puzzled about the difference in pricing, asked many people, and got the following answer: It's a matter of marketing. If all the disks will be expensive, no home users will buy them. If all the disks will be cheap, you will lose the big money (corporations and other rich customers who are willing to pay thousands for one disk). The way to eat the cake and still have it, is by selling cheap disks AND expensive disks. But how can you force the rich customers to buy the expensive disks? By using a low quality manufacturing for the cheap disks. To help people know what disk should each buy, all the IDE disks are manufactured in low quality, and all the SCSI disks in high quality. Well, it's an chicken and egg problem: to manufacture something with a higher quality, one does need to invest more, even if the underlying technology is the same. What does make adifference in IDE drives for your opinion ? And I know what - the engine . Look at Quantum Fireball CX drive and which engine it uses and at Quantum Fireeball CR - and feel the difference . There are also some differences in mechanics but they're minor . Alas, not all. I also used to think so, and it struck me already twice. I do suspect, though, that at least in one of the two cases it was just a plain lie of the Israeli reseller (Mediatek) - when it came to replacement; but does it help? :^). Notice that nowadays many HD resellers stamp on disks the _start_ of the warranty period, not the _end_, and, after 2-3 years have lasted, it's quite difficult to prove that it was promised to have a 5 year warranty. Here I agree with you. MOST of Israeli importers are *lying* to their customers in 70% of cases or trying to make money from nothing . Do you know why for example Official Distributor of WDC (I'm talking about EIM now) REFUSES to replace WD drives in the warranty period not purchased DIRECTLY from them ? This is violation of their agreement with WDC Germany but since there is NO law in Israel which protects you from this , they will piss on you and will not replace the drive until you'll begin your own revolution against them and write a bunch of e-mails to the Germany (WDC Germany) asking from them to revoke the rights from EIM . I'm talking seriously, from my friend's experince who were trying to
Re: Experience with 18G IDE HD
On Mon, Dec 20, 1999 at 05:06:38PM +0200, Evgeny Stambulchik wrote: Mike Londarenko [EMAIL PROTECTED] wrote: On Fri, Dec 17, 1999 at 01:51:54AM +0200, Evgeny Stambulchik wrote: As far as performance is considered, it's not RPM that usually matters (inspite of a big hype), but the amount of cache on disk. Until recently (1-2 years), the maximum cache on IDE disks was 128KB, while about each SCSI disk had 0.5MB. This is not exactly the truth ... [a lot of speculations skipped] Please read the whole thread (to the end); you're replying to a message in the middle. This is exactly what I did, this is why I'm replying only to your message :) Let me repeat here again (uff...): the role of cache is NOT only in prefetching data; a more import one, especially in the multitasking/heavy-load enviroments is BUFFERING of data, allowing OS to communicate with inherintly slow devices at high speeds, in bursts, hence I understand what you mean, but again, in the real world and in real multiuser, multitasking systems only really HUGE cache can help, 512K-2MB is near the same in terms of overall performance and as I said, since you frequently have cache misses it means that you have no benefit from using caching at all ! And one more important thing - you can't program the cache controller of drive itself - it decides what to cache and what not using its own firmware . Just for your information, somebody measured the speed of HDD while working under NT and compiling programs, running Winbecnch and other benchmark programs in two cases : 1)caching on HDD enabled 2)Caching disabled The difference was (don't forget it's relatively high-load) about 3% Why ? Because almost anytime drive's controller had to read/write data directly from plate. Of course cache "helps" for OS, because whenever it "asks" from drive to read something it can immediately switch to other task and run some process, do something else, etc. But remember that almost any OS has other tasks to do which need access to hard drive. But as need to read/write something grows - the benefit from cache will be smaller than from rotating speed . So, the real benefit will be only from faster (mechanically ) drive, and not from one with bigger cache . Some tests made under NT (if you wish to count this OS as *real* one :) can be found on http://ixbt.stack.net in the Storage section (Russian only) . Other similar tests were published a while ago at storagereview . Look for tests of drives with similar parameters except the RPM and cache and you will see the difference . letting it (OS) do other tasks in periods between the bursts. This is why new UARTs (16550A and up) use larger cache (and let you use _115200_ baud rate to communicate with a _33600_ modem - and, BTW, think why with DOS/Win you can do it sometimes even with a 16450 UART (1k buffer), and never under Linux), and this is why new smart bus-mastering NICs use larger buffers. Or do you believe your modem can "read ahead" from N^HBezeq, or NIC - from Internet? :) "Lama hainternet sheli lo oved ? " : (A question asked by someone who called to Netvision's tech support) What does make adifference in IDE drives for your opinion ? What it has to do with the current thread? Eli said the only difference between IDE and SCSI is the manufacturing quality and so the price tag is not backed up by any real factor, and I argued that even maintaing a higher quality itself (let alone other differences) does involve $$. So what's your point? My point is , that there is a major difference between SCSIS and IDE drives as well as between cheap IDE drives and good IDE ones :) And the difference is fortunately mechanics, the engines (motors) used in SCSIS and IDEs are not the same . I mean here the motors used to move heads and the motors rotating plates . This is not simply mechanical part, this is micromechanical part, one such motor for moving the heads sometimes can cost up to $30 (including positioning system) . Now compare this to the total cost of some hard drives and think by yourself, why SCSIs are more expensive . Another important thing is quality of plates themselves . Or , more accurately, the tests they have passed . For example there is some test which can show you how long time will pass until the first bad block will be encountered on the plate (in normal working condition). There is another one which check how plate likes hot (many drives got covered by bad-block plants just because they were intensively used in bad thermal conditions) etc, etc. Now to the Eli's speculations . For some manufacturers all this - same story and really their IDE and SCSI drives costs the same money for them , but they can't just broke the competition by lowering prices - since it will return to them as boomerang, because at some stage they will want to improve their manufacturing facilities and
