IDE HDD CF/SSD storage benchmark in a X30 **PIC**

[sjthinkpader]

I saw a lot of differ storage solutions discussed on this forum so I decided to do an almost apples to apples comparison with different types of HDD and one flavor of CF/SSD. All this was done very low cost so I don't have access to the relatively expensive real SSD or very fast CF cards.

The test platform is a 1.03Ghz X30 but loaded with 1GB of SDR RAM.

Test software is Crystal DiskMark 2.1:
http://crystalmark.info/software/Crysta ... dex-e.html



The Hitachi DJSA-220 is a year 2000 vintage drive. It performed pretty well for its age. When new drive is fragmented, performance is almost same as this drive.

The Hitachi ZIF drive is a pretty modern drive. Even with its power saving features, it can out perform the DJSA-220 even at the same rotational speed. The adapter is a $4.95 free S/H item off eBay.

The Fujitsu performance is obviously the best. But interesting that it is still slower than CF random read speeds.

The Kingston CF clearly beat the HDD's read speed due to very low latency. The sequential write speed met test conducted by Rob Galbraith in SLR cameras. While the read speed exceeded 20MB (133x 150KB) spec, write speed fell short and only yielded 54x.

Rob Galbraith's Speed Test

Random write large and small file size fell way short of the HDDs and render the X30 painfully slow whenever it need to do sizable write tasks. This is particularly a problem when the disk is almost full. I believe this is related to the erase/reclaim operation taking the same amount of time for large or small blocks of internal memory. Blank flash cells initially starts as all ones. When card level controller writes 010 to the cells, it writes 0-blank-0. Once 0 is written into any cell, only way to return the cell to 1 is with a erase. Erase is a relatively long operation. Hence writing to an old card will become slower and slower. I am interested to see if someone had ran a benchmark on a fast card such as Sandisk Extreme and Transcend 300x both new and after 6 months of use.

Sophisticated SSDs have background erase and reclaim working without user intervention. I am interested to see if low cost SSD such as Runcore can match up to Toshiba, Sandisk (M-System) and Samsung SSDs.

The CF/IDE adapter I used is one of three no name I picked up at Akihabara in Tokyo. That little shop has ZIF adapters also. The boot drive (CF) must be formatted in NTFS in order to boot. This is one way to get around the fixed vs remobable ID bit in the CIS (card information system). A RiDATA 8GB 233X card is in use in a X40 but not available to participate in this test.

These are the four devices:



Note from Admin: Images >50k.
Edit: resized images.

[x412008]

Hi sjthinkpader. Thanks for the storage benchmarks Do you have some more background info / links about the erase/reclaim operation on flash/ssd's ?

[sjthinkpader]

These are very sensitive issues for vendors. So searches yielded very little. Many patent links popped up and no doubt many trade secrets.

NAND MLC flash erase life are usually only about 10K cycles or so. Card level controllers have wear leveling, spare row, spare block replacement schemes to increase that.

What I stated is only my own experience working on flash years ago. If you look at the 4K and 512K write data, 4K write is about 100x longer. This is because it took about the same amount of time to do write or erase+write since the whole row must be written regardless of the size of file. This gets worst as capacity size of the chip became larger and the row size is longer.

[x412008]

Thanks for the info sjthinkpader

[virge]

This is great information. I've always wondered how the cheaper CF cards compared to older hard drives. Thanks sjthinkpader.

[sjthinkpader]

..... I am interested to see if someone had ran a benchmark on a fast card such as Sandisk Extreme and Transcend 300x both new and after 6 months of use.

Sophisticated SSDs have background erase and reclaim working without user intervention. I am interested to see if low cost SSD such as Runcore can match up to Toshiba, Sandisk (M-System) and Samsung SSDs.

....

Anyone using CF or SSD and has new and old comparative performance data.

[tim S]

Here are the results of the test I ran this morning.
I used the same software, Crystal Mark.
The CF card is a 16GB San Disk Extreme III.
Its results are 30.69 and 20.32 (SEQ)
30.43 and 3.882 (512)
6.523 and 0.034 (4K)
The adapter is an Addonics dual slot.
There is a 1 GB San Disk (Non Extreme) in the second slot.
Its results are 15.29 and 16.00 (SEQ)
15.25 and 8.819 (512)
5.946 and 0.137 (4K)

Tim S

[sjthinkpader]

How long have these two cards been in use?

[tim S]

The 16GB has been in use about 3 months as has the 1GB. The 1GB is just storage and is the only (San Disk) card I have that works in the 2nd slot. Both disks were used in a digital camera for a short while as well.
But neither get a full work out since I swap them with Ubuntu on the original 40GB about 50% of the time.
Tim S

[uux]

Mtron Mobi 3000 1.8" 32GB SSD in an IBM ThinkPad X40 w/ZIF adapter. Test software is bonnie under OpenBSD/i386 (sorry, no Windows). Drive is less than a week old.

Code: Select all

              -------Sequential Output-------- ---Sequential Input-- --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
X40      1000 62548 76.7 63293 23.4 31558  6.7 45824 54.5 72974 14.1 2268.8  7.1

[sjthinkpader]

For a more meaningful comparison, I suggest to use CrystalMark test software.

[uux]

Sorry, as previously stated, no Windows. I'm happy to participate if you have a recommendation of software for Unix or Unix-like operating systems that provides you with the information that you're looking for.

[sjthinkpader]

Sorry, as previously stated, no Windows. I'm happy to participate if you have a recommendation of software for Unix or Unix-like operating systems that provides you with the information that you're looking for.

Understand, the devices in my original test had OS image made only for the test. I did it that way to get as close to an apples to apples comparison as I can.

[uux]

My bad. I made the assumption from your post that you just had the general interest in the performance degrading. Also, the variations I see in hardware from the other posters, I didn't realize it was an apples to apples comparison. My usage of this hardware or software will not change in 6 months time. Any performance difference will be apparent with my previous post. If you're looking for general benchmark results, then no, I don't meet your criteria and I'm not in a position to experiment.

[sjthinkpader]

Here are the results of the test I ran this morning.
...

Tim S

Thank you, here is the updated chart:



Talked to Kingston tech support. They requested return and exchange of this 8GB 133x card. I will rerun the test on this new card and report later.

[sjthinkpader]

Added performance data from a Hitachi 7200RPM drive

Kingston return a new card of the same type and results were almost the same as before.

[T39andcounting]

Can you confirm the model number on the Hitachi? I googled it, but got no hits!

[sjthinkpader]

Sorry, the correct model number is:

7K100
HTS721080G9AT00

[sugo]

Here are results from my $65 32GB IDE SSD used in T42 running in UDMA 5 mode:

--------------------------------------------------
CrystalDiskMark 2.1 (C) 2007-2008 hiyohiyo
Crystal Dew World : http://crystalmark.info/
--------------------------------------------------

Sequential Read : 77.741 MB/s
Sequential Write : 45.696 MB/s
Random Read 512KB : 75.277 MB/s
Random Write 512KB : 31.946 MB/s
Random Read 4KB : 12.553 MB/s
Random Write 4KB : 2.138 MB/s

Test Size : 100 MB
Date : 2009/02/07 23:28:41

I have been using the drive for a while. It seems stable so far. I bought it not really for performance but for making the T42 absolute silence. I am pretty happy with the purchase.

[T39andcounting]

Any thoughts on power consumption and noise? I would assume that the faster disk drives would use a lot more power compared to solid state, and the faster the drives, the noisier?

[sjthinkpader]

Here are results from my $65 32GB IDE SSD used in T42 running in UDMA 5 mode:
....

These are great results. Can you post the brand name and model number? Is it a 2.5 or 1.8 inch form factor?

[sjthinkpader]

Any thoughts on power consumption and noise? I would assume that the faster disk drives would use a lot more power compared to solid state, and the faster the drives, the noisier?

I would think the SSD is lower in power since it doesn't have to spin the platters. I normally set the HDD off timer in power manager to about 10 minutes.

The Fujitsu's are very quiet. The DJSA-220 is very noisy. The Hitachi 7K100 are not too bad. The Hitachi C4K60 is very quiet spinning but seek has a loud click to it.

[sugo]

It's a 2.5 inch IDE drive. This is the model I am using in my T42:
http://www.transcendstore.com/ts32gssd25-m.html

Be careful that the older generation of Transcend MLC SSD gave very poor performance - there were plenty of negative reviews at newegg. The one above is the newer generation. They didn't change the model number which makes things confusing.

Regarding power consumption, at idle it's about the same as the Samsung HDD I had. At load it consumes at least 1W less.

[sjthinkpader]

Here is the updated chart:

[miggy]

It's a 2.5 inch IDE drive. This is the model I am using in my T42:
http://www.transcendstore.com/ts32gssd25-m.html

Whoa... that's impressive.
Would you mind posting some additional information about your drive? S/N and mfc date?
Or just give a hint how to distinguish between the "bad" TS32GSSD25-M and the "good" TS32GSSD25-M ?

Tommorow I'll post benchmark results of ST9160821A (ST Momentus/5400) on X32.

[sugo]

The old generation looks like this:
http://www.provantage.com/fullsize/TRSD09T.jpg

The new generation looks like this.
http://www.refbuy.com/images/TS32GSSD25-M-unit.jpg


Unfortunately they didn't really put a manufacturing date on the drive label.

I was also asked where I got it for $65, I got it from geeks.com but it seems they no longer carries the item.

[sjthinkpader]

I got one from Newegg.com for $75. If it is the old one, I'll return it.

[miggy]

ST9160821A
Seagate Momentus 5400.3 160GB ATA
IBM ThinkPad X32 (UDMA-5)

--------------------------------------------------
CrystalDiskMark 2.2 (C) 2007-2008 hiyohiyo
Crystal Dew World : http://crystalmark.info/
--------------------------------------------------

Sequential Read : 44.437 MB/s
Sequential Write : 43.612 MB/s
Random Read 512KB : 26.342 MB/s
Random Write 512KB : 27.165 MB/s
Random Read 4KB : 0.490 MB/s
Random Write 4KB : 0.774 MB/s

Test Size : 100 MB
Date : 2009/02/09 11:22:13

[sjthinkpader]

Here is the updated list:



I welcome any results from 1.8 inch solutions.

To Moderator:
This file was 29KB on the hosting site but somehow grew after posting here to 59KB. What do I need to do to maintain the same size?

[godling]

I got one from Newegg.com for $75. If it is the old one, I'll return it.

Chances are, that's an old one. I'm surprised that an MLC SSD is performing as well as that Transcend is ... for $130 (off their website) I might have to get one. The MLC stands for Multi-Level Cells and the SLC is Single Level Cells. The SLC drives perform significantly better for Operating System disks, as the random write speed is vastly superior, and is what happens (under Windows) quite often, particularly during installs.

Have you performed an install w/ that disk, or are you simply doing a file transfer?

Is use as a boot disk factored into this test setup, or simply the concept of moving data back and forth on the disk?

-Jon