michaelsil1

The Newest SCSI's

Recommended Posts

Chipstone, you have no clue what you are talking about. Your posts are full of hand waving and excitement, unaccompanied by a single fact. I suggest you reread my post. I don't think you managed to get a single thing out of it, which is unfortunate.

Olaf van der Spek already highlighted the line that demonstrates most significantly that you don't have the most elemental grasp of how a hard disk operates:

on a 15K 73GB the sustained transfer rate begins to drop as it approaches the outer limits of the disk.

I guess you did try and slip a fact in there. In retrospect, maybe you should continue to leave them out...

Additionally, if you did know anything about hard disks, then you'd know that short stroking is a common performance-increasing technique. In no way does it have anything to do with SAS. It has just as much to do with Ultra20 SCSI. Or PIO 1 ;). (Olaf also pointed this out, but your emphasis on it highlights your ignorance of the general subject we are discussing, and needs to be emphasized.)

Now, your innaccuracy on both those facts is probably obvious to just about everyone on this board. However, what follows is particularly deceitful, specifically because it is a nasty half truth, combined, in the manner that appears to be your style, with a good dose of unnecessary rhetoric:

Most of the discussions on SAS seem to proceed for an incorrect assumption; that the interface is something that is been out for a while and now being modified for SCSI drives interface.

I believe, if you read the SAS interface; ratified within the last two years, undertakes entirely new sets of protocols and interfaces. Not in the realm of U320, but an entirely new sets of benchmark performance.

I've already addressed this point. And I gave details, not a big pile of bull and rhetoric without any genuinely valuable information. Here, I'll repeat myself, and I'll make things even more obvious:

Protocol differences between SAS & SCSI:

1. Serial ATA Tunneling Protocol. This is a new addition. As its name suggests it allows SATA commands to be tunnelled through the SCSI protocol so that SATA devices can be attached to the bus. Obviously, it has no effect on the performance of SAS disks.

2. Serial Management Protocol. This protocol is also a new addition. It manages port expanders on the bus. Port expanders are a new feature in SAS, relative to SCSI 3. It also has no effect on the performance of SAS disks.

3. The device address is now larger to support more devices. This has no effect on performance, except to the extent that it allows more drives to be connected to the bus, which I already mentioned...

Protocol Similarities between SAS & SCSI:

1. Everything important for SCSI disks (our subject). Specifically, the protocols used to transfer commands and data between the controller and the disk drives are SPC-3 and SBC-3 (from my previous post). These are the exact protocols used in Ultra320 SCSI. Obviously, there will be no performance increases related to the protocol, because, well, the damn thing hasn't changed!

Maybe you need to read about the interface? You should at least stop taking press releases and general FAQs as gospel. I asked you for specifics last time, and these posts are what you give me! Not a useful detail in the lot of them!

By the way, you mention that you're an analyst, chipstone. No wonder you don't know what you're talking about. (Please forgive me, but analysts... come on, has anyone ever met one that actually knew what they were talking about? Or at least one that didn't bury you in an avalanche of regurgitated marketing bullshit --I mean rhetoric ;)...)

Share this post


Link to post
Share on other sites

By the way, Olaf,

Thank you for clarifying that ridiculously misquoted usage of my words... Chipstone, you sound like you're in politics.

Share this post


Link to post
Share on other sites

Can't we all just get along? Let's treat Chip like a human and not our doormat, friends.

kumba ya my fiends

kumba ya

I don't see what the point of specialized firmware for short stroking. Does it save time when setting up LUNS? I really don't see a point. Someone can easily write a script to make partition 1/2 a LUN then RAID those LUNS together without paying for extra firmware.

*shrug*

just my 2 cents

Share this post


Link to post
Share on other sites

Gilbo, I am not trying to make hay out of a discussion relating to a SCSI interface.

However you seem to have taken the entire discussion personally, as opposed to a professional demeanor in discussing the issue, and in the process make additional assumptions on other people’s ideas.

This is a forum for discussion, not personal attack as to the integrity of a member, and in the process, however inadvertent, insult forum members.

You must have very little in the way of manners, and no respect to speak of. Your rehearsed writing is so transparent that leave very little in the way of a viable comment.

But a great writer once said “never argue with an idiot, people may not be able to tell the differenceâ€.

It only about a drive interface; not your cosmic interpretation into Quantum Mechanics.

Share this post


Link to post
Share on other sites

Can't wait to get home and read this whole thing, because oviously someone miss understood something. Anyways, Parallel can offer higher Bandwidth but it is not efficient as Serial is.

Share this post


Link to post
Share on other sites
However you seem to have taken the entire discussion personally, as opposed to a professional demeanor in discussing the issue, and in the process make additional assumptions on other people’s ideas.

Who else can we blame for your mistakes?

Share this post


Link to post
Share on other sites
This is a preliminary HD Tach Benchmark performed on the drives being tested.

Bruno Tomagachi

The chart seems to indicate that this is a raid 0 set being tested, but with no indication of how many drives. Without knowing how many drives, the information is meaningless.

Share this post


Link to post
Share on other sites

Like I’ve tried to explain, I am unable to provide information on an ongoing Beta testing project, this chart was provided as a big favor to demonstrate that there are drives already out there, maybe not available in the retail market, but capable of exceeding the U320 threshold.

Instead of having an open mind to the fact that emerging technology may already be out there to surpass current SCSI interface speeds; a logical progression in the rapid development of new technologies, decided instead to dismiss the information, and in the process met with ridicule.

I do not have additional information. Due to confidentially agreements I was not even provided with information on controller cards, drives or testing parameters. In the near future, I hope I will be able to have more concise data to disseminate.

Share this post


Link to post
Share on other sites

Chipstone, the problem here is you fail to understand what people are trying to tell you. SAS is just an interface. Yes the INTERFACE will potentially out perform the u320 interface, but a single drive's performance will not improve due to SAS.

You started this discussion like SAS was going to revolutionize hard drive performance, when it just an interface and will improve interface / RAID array... performance. So please try to understand what people are saying. There are plenty of people on here who are more familiar with SAS than you.

Here's some reading for you if you want to learn more:

T10.org spec drafts

Share this post


Link to post
Share on other sites

Oh my. You REALLY don't have a clue what you're talking about, do you? There isn't a single hard disk drive with an STR of even 100 MB/s. The graph you gave is meaningless without a detailed config. All this shows is that there are interfaces faster than U320. I have no doubt that an array of fairly old 10k drives can provide STR of more than 320 MB/s - as long as you have enough of those drives.

You just don't seem able to grasp the difference between a HARD DISK DRIVE, an ARRAY OF DRIVES and a HOST/RAID ADAPTER.

Share this post


Link to post
Share on other sites

SAS's point-to-point, full duplex architecture and 3.0 Gb/s transfer speed coupled with Cheetah 15K.4 I/O performance enable the fastest and lowest cost of ownership storage solutions.

This is a question about the above statement:

Is the 15K3 full duplex if not wouldn't full duplex speed up the overall transfer speed or overall performance of a single drive?

Share this post


Link to post
Share on other sites

Like I tried to explain, I was provided this information, I am not a party to, or involved in the testing of SAS/SCSI drives. I was making a point based on information provided to me from IT individual that works with storage deployment, and the testing of new technologies, and also posted information form a manufacturer’s website.

I was not asserting anywhere that I personally worked with SAS, and on have an insightful personal experience with the technology, only what I’ve read, and what I’ve been told.

Hopefully this may concludes the protracted discussion on the issue until we have more tangible data on the subject.

Share this post


Link to post
Share on other sites
SAS's point-to-point, full duplex architecture and 3.0 Gb/s transfer speed coupled with Cheetah 15K.4 I/O performance enable the fastest and lowest cost of ownership storage solutions.

This is a question about the above statement:

Is the 15K3 full duplex if not wouldn't full duplex speed up the overall transfer speed or overall performance of a single drive?

Change wouldn't to would.

Share this post


Link to post
Share on other sites

I wasn't taking anything personally, chipstone, don't worry about it.

If someone makes statements that are blatantly wrong or misleading, and they're trying to pretend they know what they're talking about, when they don't, well, I'm of the belief that someone should step up and call bullshit. I apologize if my analyst joke offended you, but your style of posting makes you sound like you're trying to pull the wool over everyone's eyes. The fact is that I've contributed more genuinely useful information to this thread than your vague claims of a 'new benchmark for performance' have.

I think you should read my points and make a reply if you're interested in a discussion, as you claim to be. All you've done so far is ignore the details of the exchange and offer sweeping generalizations. Your statements are still wrong and I think you should acknowledge that or explain why you think you're right, with consideration to the points I've brought.

Share this post


Link to post
Share on other sites
SAS's point-to-point, full duplex architecture and 3.0 Gb/s transfer speed coupled with Cheetah 15K.4 I/O performance enable the fastest and lowest cost of ownership storage solutions.

This is a question about the above statement:

Is the 15K3 full duplex if not wouldn't full duplex speed up the overall transfer speed or overall performance of a single drive?

Don't see why duplex transfers would make any difference, obviously a HD is not a duplex device, either it can read, or it can write, it can't do both at the same time. I assume they mean the interface can send and receive commands at the same time, but with command queueing already built into SCSI I don't see how this would impact performance at all.

Also, it should be noted that first generation SAS has a theortical 300MB/s throughput per channel, making it theoretically slower per channel than U320 SCSI, not faster. Unlike 1st gen SATA, SAS will allow port multipliers which will making exceeding 300MB/s on a channel possible. The point-to-point star topography of SAS won't really make any difference for home users, as a standard affair dual channel U320 controller already far exceeds common sense storage setups. Unlike with ATA, where the 4 drive limit of on board dual channel P-ATA controllers usually limited to 100MB/s per channel actually was constraining to some. Having 8-10 SAS ports wouldn't provide any benefit to any of us. SAS is the future of SCSI, but just like SATA, besides the small cable, it will provide zero benefit initially no matter what kind of benchmarks you try to dig up to prove otherwise.

Gilbo, you do sound like you are taking it personally. Lay off the guy a bit. Yes, he is clueless, and has demonstrated a complete lack of knowledge in the field, but there is no reason to publically undress him by being so confrontational. We were all morons at one point, SR has is one of the most troll/flame war free major boards on the internet. Lets keep it that way.

Share this post


Link to post
Share on other sites

All specifics aside, saying "You don't know what you're talking about" isn't a very productive way to focus attention on the content of a post, and it is bordering on personal insult, which as we all know tends to make a thread less productive, not more, and can lead to run-on sentences, like this one.

This is not the kind of thing people come to SR for, whether or not their information is flawed. Not that I am one to talk, as I have certainly snapped at people in a much more insulting manner than has been shown in this thread, but I always end up regretting it.

Share this post


Link to post
Share on other sites
Don't see why duplex transfers would make any difference, obviously a HD is not a duplex device, either it can read, or it can write, it can't do both at the same time.  I assume they mean the interface can send and receive commands at the same time, but with command queueing already built into SCSI I don't see how this would impact performance at all.

This was confusing to me having dual ports on a drive to allow for simultaneous I/O; as you said a drive is either reading or writing not both at the same time. As Gilbo said I should stop falling for the media hype.

Share this post


Link to post
Share on other sites
Also, it should be noted that first generation SAS has a theortical 300MB/s throughput per channel, making it theoretically slower per channel than U320 SCSI, not faster.  Unlike 1st gen SATA, SAS will allow port multipliers which will making exceeding 300MB/s on a channel possible.  The point-to-point star topography of SAS won't really make any difference for home users, as a standard affair dual channel U320 controller already far exceeds common sense storage setups.

Duplex, is that full- or half-duplex?

And if each channel is limited to 300 mbyte/s, how can port expanders be used to lift that limit?

After all, isn't the connection between the controll and the port multiplier just a 'normal' channel that's also limited to 300 mbyte/s?

Share this post


Link to post
Share on other sites
Also, it should be noted that first generation SAS has a theortical 300MB/s throughput per channel, making it theoretically slower per channel than U320 SCSI, not faster.  Unlike 1st gen SATA, SAS will allow port multipliers which will making exceeding 300MB/s on a channel possible.  The point-to-point star topography of SAS won't really make any difference for home users, as a standard affair dual channel U320 controller already far exceeds common sense storage setups.

Duplex, is that full- or half-duplex?

And if each channel is limited to 300 mbyte/s, how can port expanders be used to lift that limit?

After all, isn't the connection between the controll and the port multiplier just a 'normal' channel that's also limited to 300 mbyte/s?

Seagate says full duplex. :D

Share this post


Link to post
Share on other sites
Also, it should be noted that first generation SAS has a theortical 300MB/s throughput per channel, making it theoretically slower per channel than U320 SCSI, not faster.  Unlike 1st gen SATA, SAS will allow port multipliers which will making exceeding 300MB/s on a channel possible.  The point-to-point star topography of SAS won't really make any difference for home users, as a standard affair dual channel U320 controller already far exceeds common sense storage setups.

Duplex, is that full- or half-duplex?

And if each channel is limited to 300 mbyte/s, how can port expanders be used to lift that limit?

After all, isn't the connection between the controll and the port multiplier just a 'normal' channel that's also limited to 300 mbyte/s?

Seagate says full duplex. :D

Maxtor also claims simultaneous read writes.

Share this post


Link to post
Share on other sites
And if each channel is limited to 300 mbyte/s, how can port expanders be used to lift that limit?

After all, isn't the connection between the controll and the port multiplier just a 'normal' channel that's also limited to 300 mbyte/s?

Port expanders do NOT increase the channel bandwidth.

You can get from 3-4 drives on a 300 MB/s port, which would give each drive 75-100 MB/s - overhead. If you using the new 15k drives and are aiming for maximum performance you may only choose to use 2-3 drives per SAS port.

Share this post


Link to post
Share on other sites
Also, it should be noted that first generation SAS has a theortical 300MB/s throughput per channel, making it theoretically slower per channel than U320 SCSI, not faster.  Unlike 1st gen SATA, SAS will allow port multipliers which will making exceeding 300MB/s on a channel possible.  The point-to-point star topography of SAS won't really make any difference for home users, as a standard affair dual channel U320 controller already far exceeds common sense storage setups.

Duplex, is that full- or half-duplex?

And if each channel is limited to 300 mbyte/s, how can port expanders be used to lift that limit?

After all, isn't the connection between the controll and the port multiplier just a 'normal' channel that's also limited to 300 mbyte/s?

Seagate says full duplex. :D

Seagate didn't claim their drives were duplex, but that the interface is, and it just occurred to me what they were talking about. By claiming to be full duplex I believe they mean that each channel can send and receive data at 3Gb/s simultaneously giving a max theoretical duplex throughput of 6Gb/s. Parallel SCSI/ATA can only send data in one direction at a time, so U320 is limited under any circumstances to 320MB/s. If you had 2 large arrays on the same channel, one sending data to the other, that would be beneficial, but for one drive or any more mundane setup, this wouldn't seem to be of much use.

Then I read Maxtor's release you posted, and I have no idea what they are talking about. Unless they are talking about reading a writing to the drive's buffer, I have no idea how a drive can read and write to the media at the same time unless Maxtor has started using more than one read/write head at once which wouldn't be very practical or usefull if the heads are still all on the same assembly.

Olaf van der Spek, you misunderstood me. Port expanders won't allow you to increase the bandwidth of a SAS channel. What I meant, was that port expanders will allow you to connect enough drives that could potentially exceed the available bandwidth and make it a real bottleneck. With current SATA, when you connect a SATA drive to a channel, that's all you can connect, leaving the remaining 80-90MB/s of bandwidth wasted, so increasing the bandwidth from PATA 133MB/s to 150MB/s SATA was worthless. With SAS, all available bandwidth will be able to be utilized.

Share this post


Link to post
Share on other sites
You can get from 3-4 drives on a 300 MB/s port, which would give each drive 75-100 MB/s - overhead.  If you using the new 15k drives and are aiming for maximum performance you may only choose to use 2-3 drives per SAS port.

No, if you require max STR you need to limit the number of drives per channel.

Note that max performance doesn't equal max STR in most situations.

In random IO scenarios like DB servers, even 16 drives on a channel probably won't saturate the channel.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now