X-bit labs

Started by: tygrus | Date 04/23/08

Comments: 19 | Last Comment:  05/02/08

[1-16]

1. Most IO graphs would be better reported as MB/s and provide table to display results. The graphs you have are only about 10% useable. IOps are only of interest at the extremes of max throughput, max IOps and min IOps across all block sizes (not per block).

Sometimes several graphs from IOmeter become redundant when they are highly correlated (same result presented three different ways eg. IOps, MB/s, ms/IO).

FLASH based SSD require more/expensive components to reach higher sequential and random read and especially write performance. FLASH based SSD are not as good as they could be.

I would also like to mention here that many of the graphs would have been better if they'd been presented just as they were, only plotted on a logarithmic, rather than linear, scale. This would prevent the outliers from compressing the rest of the data into a uselessly small range.

Other than this relatively minor criticism, I find it hard to find any fault with this review. Very thorough method, and conclusions very well backed by the results. Another excellent article from X-bit.

2. That I-Ram rocks! Ideal for memory files!!!

3. Great review! I enjoyed every bit of it.
Can you do a review of the FusionIO drive and compare it to these drives when it comes out? I would love to see how it performs against the i-RAM.

4. Btw, Synthetic benchmarks are total crap. I set up RAID, and in all synthetic benchmarks its 2x faster than single HDD. However when testing in real applications including Office, C++ compilers and linkers, etc. RAID is often slower, and never faster than 3%. I really dont understand companies which make these crappy "benchmarks". They are totaly misguiding. Worst thing is that they have "Office" benchmark, "Workstation" benchmark, etc. and if you test in exactly same real world applications (Office, Workstation) you dont get even near results. Only thing I did not test is Database servers, I dont know how much synthetic benchs lie there.

5. Interesting review. Would be quite interested in seeing what the same benchmarking procedure gives using file systems suited to SSD : Isn't it weird to use a SSD with a file system engineered to optimize read/write on rotating platters while there is no such things on SSD. see :
http://sourceware.org/jffs2/jffs2-html/
http://osl.sed.hu/~havasi/ubifs/#background

6. Is there anything like comparable to iram that will handle 8 or 16gb?

That would be so good for premier scratch disks or stream audio samples for a live PA.

Yes, the HyperDrive4 is the fastest consumer storage available. http://www.hyperossystems.co.uk/. 16GB per card, and they sell it in 4x raid boxes for 64GB all the way up to 24x 384GB!

7. darn... drooollll.... where can i get a hold of that i-RAM.......

8. there is an I-ram 2.0, its sata2.0 and uses ddr2 memory and is installed as a bay-device, like a cd-rom drive, just look at ebay

i thought so as well

supports more then 4gb of ram as well

9. actually, i am completely wrong, its still sata and ddr... =(

get to it, gigabyte

10. Didn't throw the 150GB Raptor into the fray? You guys didn't get an eng sample of the new VelociRAPTOR like a swarm of other sites did? I would have to think the I-ram would be useful in situations if a user is planning on playing only a certain application or game (when this first came out 4GB would have been large enough for any game) at a time. If one is all over the map in terms of application usage, it's only virtue would be for going "plaid" with a swapfile.

I can just imagine the load times saved by using one for playing WoW 16+ hours a day, but when i deleted my WoW install it was in the region of 6.5GB of space used.

I'll give it another look once a new version supports SATA II and one can jam 8 2GB (where I'm at, a local store has a 2x2GB kit for as little as 116$). If it becomes popular that could be a pretty decent option for 500$ or so in the future.

But surely I ramble on too much and it is too early to be excessively coherent!

11. if Gigabyte had an i-RAM that supported DDR2 up to 16GB's, SATA-II and cost under $400 i'd jump all over that in a heart-beat.

12. It would have been really interesting to see how the fastest SSD:s stack up against these. The Samsung SSD:s are at the low end of SSD performance. Memoright have drives that they claim can do 120/120 MB/s and random writes of 550 IOP/s.

At that speed I think they should be able to beat at least the SAS drive in all tests. Providing the manufacturer is reporting the random IOP/s correctly.

Also it would be really interesting how http://managedflash.com/newproducts/slpcsoft.htm drivers would affect these results.

13. The SSD offerings of Samsung are a joke because they protect their traditional HD business.

MTRON and OCZ make much faster drives with speeds of 120/90 and 100/80 MB/s. Newegg currently has the OCZ 64 GB 2.5" one for a rather low price of 1000$, which is CHEAP for a fast SSD. They need to start selling here because we can only get 64 GB MTRONs for 1700$ 3.5", so useless for Laptop use :(

14. We ordered 2 x Gigabyte i-RAM Box from PC Club,
Gigabyte's USA distributor; they took our deposit,
then Gigabyte took 2 months to say that they
would not be distributing the I-RAM Box in the USA.

It is available in Asia, e.g. via distributors in Hong Kong.

See our fair use collection of photos and the
i-RAM Box User Manual here:

http://www.supremelaw.org/systems/gigabyte/

So, we've filed a Provisional Application for U.S. Patent
for far superior alternative to that i-RAM Box
and to the HyperOS HyperDrive4 (a UK product).

We'll announce our progress via the Internet,
as soon as more news becomes available.

Email: supremelawfirm@gmail.com


Thanks, all!


Sincerely yours,
/s/ Paul Andrew Mitchell
Webmaster, Supreme Law Library
http://www.supremelaw.org/

15. > if Gigabyte had an i-RAM that supported DDR2 up to 16GB's, SATA-II and cost under $400 i'd jump all over that in a heart-beat.


Because SDRAM is so much faster than current SATA-II interfaces,
a superior setup -- for purposes of maximizing speed --
is to configure 2-4 in a RAID 0.

By switching to 2.5" hot-swap form factors using SO-DIMMs,
four (4) such "LapTopRAMDrives" (tm) (see patent application)
can fit into a single 5.25" drive bay, using the excellent
QuadraPack Q14 from Enhance Technology, Inc.

There are similar 4-in-1 enclosures made by Supermicro
and Addonics, and a few cheap imitations from Taiwan
sold by Newegg.

Thus, 4 @ 4GB = 16GB total, running parallel in RAID 0
which exploits 4 parallel SATA-II cables much better
than a single SATA-II cable running at only 300MB/second.

Coupled with a modern x8 lane PCI-Express RAID controller
e.g. Areca with 2GB hardware cache, now we are talking
a very fast storage subsystem (albeit space-limited).


Sincerely yours,
/s/ Paul Andrew Mitchell
Webmaster, Supreme Law Library
http://www.supremelaw.org/

16. This makes me wonder what would happen if you used a flash drive known for speed instead, and added a 32MB high-speed cache.

[1-16]