SSD Sata2 help, slow performance? [SOLVED]

petlab · Last edited by petlab on Mon Feb 09, 2009 7:43 pm; edited 1 time in total

I bought a Patriot 64GB SSD v.2 and I am not getting the bandwidth I expect, and I think my controller may be the problem. I am getting reads at around 75MB/S according to hdparm, bonnie++ says about 88MB with 18%cpu.
I am not expecting the 175/100 read/write as specified by Patriot, but I was hoping to get above 100 reads with this setup.

zyko · Posted: Sat Feb 07, 2009 11:49 pm Post subject:

Does your controller do ahci && is it enabled in your kernel?

NeddySeagoon · Posted: Sun Feb 08, 2009 12:00 am Post subject:

petlab,

If you believe

petlab · Posted: Sun Feb 08, 2009 12:50 am Post subject:

zyko: i'm looking into that, to see if it would help. Neither of these controllers does ahci, afaik. They are too old.

NeddySeagoon: Right. I am trying to figure out what those M$windows people are measuring. Using different benchmarks, they show "measurement" graphs with this drive doing 175MB/s sequential reads - somehow. Also the Intel X25s supposedly can do 225MB/s. With all these 'fast' drives around, I'm trying to see what they're measuring. It isn't totally fake, IMO.

I think it is supposed to be either burst or sustained THROUGHPUT. Either would be fine with me. AFAIK there is no way to measure or advertise 175MB/s sequential reads if a drive can only do udma100. Neither burst or sustained. So there is something I still don't know here. My drive doesn't have a buffer, so it can't be sata bus only measurements.

I have spent ~8 hours surfing for a quick answer, but it seems to be a hole in cyberspace. Lots of people linking to t13.org, intel, yadda yadda. I need the end-user answer, not the 10k pages of specs. After a while, if nobody answers, I'm going to bother Jeff Garzik(?) and ask what "225MB/s" means.

I am aware of the general lack of knowledge around bus speed, thruput, dma, etc., and don't need to talk those over. We'll see what is faster than udma6 and beyond.

I'll let us know what I find out.
_________________
Get Serious - Get JAWA CZ

NeddySeagoon · Posted: Sun Feb 08, 2009 10:58 am Post subject:

petlab,

I was trying to make two points.
1. The drive may or may not do better than the 100Mb/sec
2. Your PCI card won't let you go any faster anyway.

To go faster, you need a SATA2 controller on the motherboard or in a PCIe or PCIx slot.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

petlab · Posted: Sun Feb 08, 2009 7:45 pm Post subject:

NeddySeagoon: If I seemed rude, I apoligize. I wasn't meaning to be curt. I agree with 1. and 2. It seems the PCI slot is running at 66MHz, so possibly it could transfer what, like 233MB/s theoretical. I know that isn't real-world.

Here is the part that doesn't make sense: My friend has this same drive, on a much faster machine. Using bonnie++ I show sequential reads between 72 and 91 MB/s. Wow! But his system reads 139MB/s!

I don't understand how udma5 (100) could do that. I'd like to know why the speeds can be above 100. I figure bonnie++ is relatively accurate, so hmmm. Still learning here. I have a post on Patriotmemory.com asking about this, hopefully I will get a response Monday. I will check my friend's drive too, to see what it reports for udma.

Thanks again for your interest and help, I hope to make this clearer for all.

petlab

NeddySeagoon · Posted: Sun Feb 08, 2009 8:49 pm Post subject:

petlab,

Its not rude - there is nothing to apologise for.

I'm not sure I believe the

petlab · Posted: Sun Feb 08, 2009 9:00 pm Post subject:

I do have onboard sata1 and got the same speeds. I'm looking at:

petlab · Posted: Mon Feb 09, 2009 7:16 pm Post subject:

EDIT:
I talked to my friend with the same drive last night, and came to a decision on several points after more some more google study. I am still largely ignorant in this area, so take this lightly:

UDMA/SATA - They are completely different. The udma spec defines how PATA drives can transmit data. Some older controllers can only treat SATA drives as a PATA ones, so this is included as a compatibility feature on some sata drives. Those older SATA/PATA controllers can only get 100MB/s out of the drive because that's all the controller can do.

SATA/SATAII - The libata drivers, well documented on http://linux-ata.org will handle a drive in true SATA mode, for supported chipsets. These drivers are then capable of transmitting "full rate," which is either 150MB/s or 300MB/s approximately. So this is completely different than UDMA-xx. This is how people can see sequential reads in the range of 250MB/s on Intel X-25 drives. They aren't using UDMA-anything.

AHCI/SATAx - AHCI is cool because it can do NCQ for drives that support it. NCQ is a method where requests are resequenced by both the drive and the controller for maximum throughput/ lowest latency. This is similar to what the kernel does, with the I/O schedulers you may have read about. A controller with full AHCI support can really improve bandwidth, even for SSDs. My drive and/or controller don't support AHCI, so I don't have that extra boost.

My sequential read speed vs. my friend's - My system, when in runlevel 2, shows a sequential read speed of 91MB/s. My system is a 32-bit athlonxp running at 2088MHz. When I'm in KDE4, with lots of services running, I see only 75MB/s. My friend's system is a dual socket opteron machine with each chip has 2 cores. Much faster machine, to say the least. His bonnie++ output shows 139MB/s for the same drive. I didn't ask, but I think that was taken with ALL his services running (lots). So, after checking everything, I found that the speed difference is due not to misconfiguration, but machine speed.

This is good news for me, since it means that the drive is configured properly and it can send data as fast as my poor old machine can ask for it. It also means it's time for me to get a faster machine...

Thanks for all your input
petlab
_________________
Get Serious - Get JAWA CZ