Slashdot Mirror
Seagate Firmware Performance Differences
Derkjan de Haan writes "The Seagate 7200.10 disk was the first generally available desktop drive featuring perpendicular recording for increased data density. This made higher-capacity disks with excellent performance cheaper to produce. Their sequential throughput actually exceeded that of the performance king — the Western Digital Raptor, which runs at 10,000 RPM vs. the more common 7,200 RPM. But reports began to surface on the Net claiming that some 7200.10 disks had much lower performance than other, seemingly identical disks. Attention soon focused on the firmware, designated AAK, in the lower-performing disks. Units with other firmware, AAE or AAC, performed as expected. Careful benchmarks showed very mixed results. The claims found on the Net, however, have been confirmed: the AAK disk does have a much lower throughput rate than the AAE disk. While firmware can tune various aspects of performance it is highly unusual for it to affect sequential throughput. This number is pretty much a 'fact' of the disk, and should not be affected by different firmware."
When the performance of a lower-end drive is better than that of a higher-end (or, god forbid, a SCSI drive!) this is a serious bug that of course needs to be fixed in the firmware update.
I'll take reliability over performance of a hard drive any day. Nothing sucks more than swapping out drives.
Sigh, never mind. Ubuntu's been updated since I put this computer together, so now hdparm /can/ talk to a SATA drive.
Wouldn't you know that I've got an AAK disk.
Hail Eris, full of mischief...
E pluribus sanguinem
Looks like Seagate designed the new drives for servers (probably file servers) because they're really good a moving large chunks of data around, doing large reads, and large write, but not so good a a ton of little reads and writes. So, don't buy them for your desktop/workstation.
- I voted for Nintendo and against Bush
...why they named it AAK!
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
Disks are cheap. I *always* run a RAID1 mirrored pair in my PCs, as pretty much all mobos these days have RAID1 capability built into the chipset's SATA controller anyway.
On my main machine at home, I always buy my disks in groups of three drives whenever I upgrade. Two drives stay in the machine as the mirrored pair, and once a month I pull one out and stash it in a safety deposit box at my bank, and put the third drive into the machine and re-sync the mirror. That way if my house burns down / tornado smashes it or whatever bad thing that might happen, I've got a drive with my machine's image on it, no older than one month, stashed away offsite in a secure place so I can recover most all my stuff to a new machine.
From TFA page 6:
A sad detail is that updating an AAK disk to other firmware is impossible, due to physical differences of the two disks.
(emph. mine)
Different disks have different performance. News at 11.
:wq
This number iis pretty much a 'fact' of the disk, and should not be affected by different firmware.
:)
Poor spell checking is pretty much a 'fact' of the browser you use when you submit articles to Slashdot, and should be affected by different editors.
Perhaps kdawson's firmware is broken?
Whatever you do, don't stream audio from one of the -K drives across Vista!
It's interesting to note that the general purpose benchmarks come out with AAK in the lead while the others, all very much sequential read focussed, don't. So the question is, what exactly are the operations that the AAK is doing faster in the mixed benchmarks? Seeking? Or maybe it's a bus bandwidth limit at the hard drive end?
Sadly, we can't tell, because the author has focussed on the sensationalism of poor performance rather than asking these questions. Seems to need a few more experiments setting up, or alternatively an answer from the horse's mouth.
Some candidate theories:
- microcontroller software bug (unlikely)
- hardware cost-down such as a slower, cheaper microcontroller or less RAM on the drive (quite likely)
- rebalancing the performance optimisation, changing the cacheing or readahead algorithms to suit typical loads (possible, but it seems odd that this would limit linear read performance)
I have been setting up a couple of 8-drive RAID-5 arrays with these drives for some customers, and I also found out that 3.AAE drives performed much better that 3.AAK. No idea why. Seagate was unresponsive to queries about flashing the firmware and I had to replace all the 3.AAK drives by 3.AAEs.
The manufacturing country had nothing to do with it. I had some chinese 3.AAE and 3.AAK as well as taiwanese (or was that thai?) 3.AAE and 3.AAK. 3.AAE would always perform better.
The kind of testing I performed was:
Also, if you buy a retail kit (which I found cheaper than OEM at Fry's), there is no way to find out the firmware level on the box. You had to open the retail boxes to check the firmware revision on the drive itself.
One theory I have is that these drives can supposedly be configured for server or workstation workloads. It could be that AAK drives are configured for server workloads by default (unless overridden) while the AAE are configured for workstation workloads by default. I have no idea how to toggle this under Linux.
Quite why people suddenly think that drives are going to fail catastrophically at the same time like this is be
An experienced administrator would know there is one item in the data center everything is relying on no-one could ever think of it failing, and it will fail at the most catastrophic time you think of. It won't be all fo those 1000'thns drives failing at the same time because some plane mistook your server lights for the landing runway, It will be some cheap sprinkler, the security lock of the door, Or some manager that decides to shutdown a machine to protect it from a Denial of service attack.
If there is no such item a good BOFH will create such red button.
Because you have a harddrive but no firmware.
Lars T.
To the guy who modded me down from perfect to terrible Karma - Apple haters still suck
yup, about a decade ago I worked somewhere where this was an issue - they had a RAID configuration of somekind (I'm a nerd, but not a hardware one) and they had bearing failures in sufficiently close succession that the third failure occurred before all of the swapping from the second failure hadn't been completed.
supposedly it was traced to a common fault in the bearings
Also, if you're concerned about Linux block device performance, look at the various kernel tunables. On a single drive, such as those Seagates, I can get extra ~10MB/s. On RAIDs and LVM volumes, the differences can be much higher-- more than twice as fast, in some cases. There are a few parameters that make a difference, and many values you might want to try for each. I have a script iterate through the various permutations, running IOZone on each, so I can see what does best for read vs. write and large vs. small file performance. But I can't release it just yet (employer makes 100% of income from Open Source; employer hates Open Source). Anyway, somebody out there can do better than I, I'm sure :)
This discusses the tunables you'd want to check: http://www.3ware.com/KB/article.aspx?id=11050
Note that these do NOT apply only to 3Ware controllers. And the differences in performance can be massive.
The security for the door was malfunctioning earlier this summer, and the alarm was going off. The security guard thought the button was a shutoff switch for the security system... Luckily we had redundant servers at another location... Of course half of those didn't work...
Luckily also, this was the smaller data center at that site, so it only housed a few hundred servers... including the servers that ran many of our ATMs, and our server inventory and trouble tracking software... which didn't fail over to their backups... of course.
In addition, we had no idea where the server housing our server inventory information was... It turns out it was housed on a server called Skywalker... which we couldn't find... It turned out to be a cluster of Anakin and Amidala...
Fracking geeks.
The television will not be revolutionized.
For those who are unclear on what perpendicular recording is, Hitachi made a video explaining how it works. It's a bit dry and technical, but I figure the Slashdot crowd is savvy enough to grok it.
In general, it is safe and legal to kill your children. -- POSIX Programmer's Guide
I am an insider in the drive industry, so while I need to be vague on some things, I can add clarification on others.
A hard drive is a very complex subsystem inside your computer, more complex than many people realize. A hard drive contains one or more CPUs, memory, firmware, and dedicated hardware devoted to the functions of storing and retrieving data.
There is no single "right" way to draw the line between what is firmware and what is hardware in a hard drive. Algorithms could be coded in VHDL or Verilog and synthesized into the silicon, or they could be compiled in C (or hand coded in assembly) and be embedded in firmware. Each drive company has their own philosophy for where to draw the line.
Some drive companies choose to implement only fundamental functions in silicon, and implement everything else in firmware. For these companies, comparing their firmware to the BIOS in a PC is a poor analogy. A better analogy would be to compare the firmware to the operating system.
In a system with "lite" firmware, the firmware typically would be responsible for configuring a few control registers and buffers, and then the hardware would take over. But for a system with "heavy" firmware, the firmware behaves much more like a kernel. Data is not going to be moved in or out of buffers, or be sent to and from platters, without the active involvement of the firmware scheduling and ordering that activity.
The author of the OP wrote "it is highly unusual for (firmware) to affect sequential throughput". The author is wrong. In a system with "heavy" firmware, all performance is highly dependent on the firmware. It can easily make the same difference in performance as you would see running Windows 95 v. Windows XP v. Windows Vista v. RH 7.2 v. RHEL 3.0 on the same PC hardware.
I do not know if the Seagate drive in question is a "heavy" or "lite" firmware drive, but I do know that the assumption that firmware takes a minor role in hard drive performance is mistaken.
The difference between these drives is not only the firmware, the hardware is also different. If you look a bottom of the drives, you can see the board has a completely different layout and presumably (the pictures I've seen were too low quality and the memory was not on the visible side on the AAK-drives) different chips. According to Seagate, the AAK drives were for an OEM-customer (unfortunately, they didn't mention which one). But how or why those drives made it to retail-channels (Seagate and the OEM-customer knowing the drives had a different performance profile)?
It only takes one man to change the Wisdom of the Crowd to Tyranny of the Masses.