Now? You want to do this now?
I have a right to know! I'm getting married in four hundred and thirty years!

Sunday, May 29

Geek

That Was To Be Expected

Four installs and two failed upgrades later, I'm up and running.

Yay!

Fedora Core 4 Test 3 came out on the 10th of May. Today is the 28th. There are already 1.2 gigabytes of updates available.

Also to be expected: There's more than one type of DVI cable. This is the wrong type.

Hmm. Okay, what I have is a DVI-I single-link cable. What I need is a DVI-D cable... Or a pair of needle-nose wirecutters. What I want is a DVI-D dual-link cable. Oh look, here's one, only $250. Maybe I'll try another store...

Posted by: Pixy Misa at 09:01 AM | No Comments | Add Comment | Trackbacks (Suck)
Post contains 105 words, total size 1 kb.

Geek

Push the Button!

I went and did it. Currently installing about 8GB of software. Should take about half an hour.

Oy.

Lots of shiny new toys to play with; my old Linux box was running Fedora Core 2, which is a year old now.

What's amazing is the amount of stuff that's left out. Nethack? Pike? Metakit? No sign of them, or of ten thousand other packages. Okay, 90% of them are crap, but that leaves an awful lot of good stuff.

Posted by: Pixy Misa at 04:30 AM | No Comments | Add Comment | Trackbacks (Suck)
Post contains 85 words, total size 1 kb.

Geek

Semi-Splut

So, I backed everything up yet again, checked it, double-checked it, triple checked it, and then installed the replacement video card in my new computer (the GeForce 6600 non-GT).

Stuck the CentOS 4 disk in and discovered that my computer no longer has a DVD-ROM drive. Ten minutes of fiddling with the BIOS options got it back, and it turns out that CentOS 4 has no idea what a GeForce 6600 is, and crashes when it tries to go into graphics mode.

Crap doodles!

Okay, pulled out Fedora Core 4 Test 3. That works. Mmmf. This means I can't easily do a test upgrade before wiping the LVM that it built when I gave up fiddling with that poxy pagan piece of shit.* Unless, maybe...

I'd say the hell with it, but this machine currently contains my only complete, easily accessible copy of my anime fansub collection.

* Disk Druid, Red Hat's partitioning utility. It sucks.

Posted by: Pixy Misa at 03:13 AM | No Comments | Add Comment | Trackbacks (Suck)
Post contains 158 words, total size 1 kb.

Saturday, May 28

Geek

All I Want For Christmas

Is one of these puppies. 8 64-bit MIPS cores, 32-way multithreaded (4 per core), two 64-bit DDR2-800 memory controllers and a 32-bit QDR SRAM controller, four gigabit ethernet ports and two ten-gigabit ethernet ports, HyperTransport and PCI-X.

Even better, rip out the PCI-X and replace it with twenty channels of PCI-E, which would be a huge improvement. Drop Linux or BSD on it and it will power a thousand websites without even breathing hard.

Posted by: Pixy Misa at 06:21 AM | Comments (3) | Add Comment | Trackbacks (Suck)
Post contains 84 words, total size 1 kb.

Wednesday, May 25

Geek

Well, Yeah

"It says 'Golgafrincham Ark Fleet, Ship B, Hold Seven, PHP Programmer Second Class' — and a serial number."

"A PHP programmer?" said Arthur, "a dead PHP programmer?"

"Best kind."

Posted by: Pixy Misa at 02:33 AM | Comments (3) | Add Comment | Trackbacks (Suck)
Post contains 33 words, total size 1 kb.

Tuesday, May 24

Geek

The Geek Is Strong With This One

The geeks are out in force (hee hee) at Ace's place tonight.

I know this could get me booted from the geek club (Hey, wait! I have a 64-bit computer with 1.8 terabytes of disk! Shiny, shiny!) but I have to admit that I saw Star Wars (not Episode IV, Star Wars) twice at the age of 12, and haven't seen any of the other movies.

Posted by: Pixy Misa at 11:51 AM | Comments (2) | Add Comment | Trackbacks (Suck)
Post contains 80 words, total size 1 kb.

Monday, May 23

Geek

Apart From That, Mrs. Lincoln...

This shows just what a modern PC is capable of:


avg-cpu: %user %nice %sys %iowait %idle
0.42 0.00 60.42 39.16 0.00


Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
sda 1080.00 69093.05 2.53 328192 12
sdb 1090.53 69793.68 0.00 331520 0
sdc 1110.32 71060.21 0.00 337536 0
sdd 1099.37 70359.58 0.00 334208 0
sde 661.89 42361.26 0.00 201216 0
sdf 647.58 41458.53 0.00 196928 0
sdg 277.68 17771.79 0.00 84416 0
sdh 279.16 17866.11 0.00 84864 0

Thats 400MB/sec of I/O, limited by the PCI connection to the second SATA controller. I have two PCI Express x1 slots, each of which can provide double that bandwidth, assuming I could find a cheap PCI Express SATA card (hah!) which would push my bandwidth up to 500MB/sec, which is as fast as the disks can go.

This isn't a fancy server motherboard either, just an ordinary desktop one. Albeit a nice desktop board.

This is my old system:


avg-cpu: %user %nice %sys %iowait %idle
0.40 0.00 43.40 56.00 0.20


Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
hdc 666.60 42611.20 8.00 213056 40
hdd 673.40 43046.40 8.00 215232 40
hde 445.00 28480.00 0.00 142400 0
hdf 445.00 28480.00 0.00 142400 0
hdg 503.60 32243.20 0.00 161216 0
hdh 503.80 32230.40 0.00 161152 0

Which isn't bad - about 200MB/s - but not in the same class. Here the fault is partly PCI, partly IDE, partly the previous generation disk drives.

My old 120GB IDE drives get around 45MB/s maximum transfer rate. The newer 200GB SATA drives reach 55MB/s, and the brand new 250GB drives can reach 70MB/s.

One interesting thing: The next major advance in hard disk technology is perpendicular recording, where the magnetic domains that hold the data are oriented vertically into the disk rather than longitudinally along the track. If you think of current bits like dominoes lying flat, perpendicular recording makes them stand on end, allowing them to be packed much closer together, up to ten times.

Most previous advances have come by making the overall area of the domains smaller, by making them shorter and packing the tracks closer together. Since the rotational speed of disk drives has increased only slowly, this meant that as the number of tracks increased, the time taken to read an entire disk also increased, climbing from minutes to an hour or more.

Perpendicular recording will increase transfer rates in direct proportion to the size of the disk, avoiding this problem. However, if we do achieve a ten-fold increase in capacity, transfer rates will far exceed even SATA-II's 300MB/s per channel.

Well, darn.

Posted by: Pixy Misa at 01:20 AM | Comments (1) | Add Comment | Trackbacks (Suck)
Post contains 436 words, total size 3 kb.

Sunday, May 22

Geek

Yucky Lack Of I/O Bandwidth

This is crap:

avg-cpu:  %user   %nice    %sys %iowait   %idle
0.00 0.00 27.45 72.55 0.00


Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
sda 0.00 0.00 0.00 0 0
sdb 0.00 0.00 0.00 0 0
sdc 0.00 0.00 0.00 0 0
sdd 0.00 0.00 0.00 0 0
sde 670.14 42901.80 0.00 214080 0
sdf 662.12 42375.95 0.00 211456 0
sdg 266.73 17070.94 0.00 85184 0
sdh 268.74 17199.20 0.00 85824 0

Two bad things going on here. First, the Sil3114 that provides the four SATA-I ports on my new motherboard is a PCI device, giving a maximum throughput of 133MB/sec (theoretical). It's getting 119MB/sec here, which is actually very good for PCI.

Second, although I'm running the exact same test on all four drives, the first two are giving two-and-a-half times the performance, presumably due to some sort of interrupt or DMA prioritisation.

So far this time it hasn't actually crapped out on me with a flood of spurious errors. We'll see how long that lasts.

Oh yes. I now have eight SATA drives in my new computer, a total of 1.8 terabytes of storage. This wasn't the original plan, and in fact is making things rather difficult, but I couldn't get things to work reliably in the old system.

Meh.

Posted by: Pixy Misa at 10:21 PM | No Comments | Add Comment | Trackbacks (Suck)
Post contains 218 words, total size 2 kb.

Saturday, May 21

Geek

A Bridge Chip Too Far

I was trying to do a final check to make sure that everything had been copied to the new box correctly, but after four crashes in two hours due to the flaky old SATA controller, I gave up.

Shut everything down, installed the new Promise controller, tweaked a couple of files so that it was recognised, and off we go - except the main RAID array wasn't recognised. And when I tried the old trick that recovered it last time, it wasn't having any. 600GB of toast.

Good thing I backed up all that toast first. Well, that was the whole point of the exercise. Indeed, it was the idea from the start, redundant arrays of inexpensive toast. Disks. Whatever.

I told you butter wouldn't suit the works.

Update: Splut.

This means war.

Posted by: Pixy Misa at 08:13 PM | No Comments | Add Comment | Trackbacks (Suck)
Post contains 143 words, total size 1 kb.

Friday, May 20

Geek

Bonk/Oif

Let's see, 56 GB to go at hmm, 40 MB per second, which will drop down to 30 or so by the time it gets to the inner tracks, so call it 35 on average, that's 1600 seconds, a bit under half an hour, and this particular ϖ∂* run will be over.

It's done nearly 150 GB without errors, and normally it spluts within the first few minutes.

What's the difference?

I'm only testing one drive at a time.

I've run into this once before, and it's really, really nasty: With certain revisions of certain SATA controllers, everything works perfectly if you only have one such card in the machine. Things start to get a little flaky if you install a second identical card, but you can't track it down.

But, if you run ϖ∂ or some other stress test that runs both cards flat out at the same time, you get inundated with spurious errors.

I couldn't tell that this was happening because on this particular system, the spurious errors caused a "screaming interrupt" which effectively locked up the system. On the previous system, I could at least see that running two copies of ϖ∂ at once would report huge numbers of errors, which would stop immediately if one copy of ϖ∂ was terminated.

About 34 GB to go now. I think maybe the Promise SATA150-TX4.

28 GB to go. The thing I couldn't understand was that it looked like I had two bad drives and two bad controllers, which just didn't make any sense. Bad drives are a common enough phenomenon here at Pixy Central, but two out of four brand new drives and the controller card and the motherboard chipset? Seems a bit much even given the infamous Pixy Central Entropy Field.

22 GB to go. If I order now, I should get the card for the weekend, which would be nice.

This still leaves unexplained the spluttage of the IDE raidset. I guess I'll have to chalk that one up to gremlins.

14 GB to go. Order placed. Pity I didn't work this out at the time I was building the machine. Sigh.

Test completed, no errors. That's three done. Now, /dev/sda seems to be poigled right now from previous tests, so I'm going to reboot and run the test again.

Uh-oh, no rebootee. Have to wait until I get home tonight.

* Latterly known as badblocks.

Posted by: Pixy Misa at 02:21 AM | No Comments | Add Comment | Trackbacks (Suck)
Post contains 403 words, total size 2 kb.

<< Page 1 of 3 >>
61kb generated in CPU 0.06, elapsed 0.2379 seconds.
54 queries taking 0.2042 seconds, 290 records returned.
Powered by Minx 1.1.6c-pink.
Using http / http://ai.mee.nu / 288