What is that?
It's a duck pond.
Why aren't there any ducks?
I don't know. There's never any ducks.
Then how do you know it's a duck pond?

Saturday, November 08

Geek

LiveDrive

Put the ioDrive into production today.  That's a big load of worries off my mind.  The production server (this is at my day job, not mu.nu/mee.nu) has been constantly on the edge of flattening its RAID array, which is after all only four SATA drives.

Four 7200 rpm SATA drives can only do 480 iops, best case, and the ioDrive can do about 6000 iops, worst case, so it's making things run a lot more smoothly.

Now if I can just get the query builder to stop selecting the wrong index and sorting 15 million records every time a user does a search, everything will be peachy!

Posted by: Pixy Misa at 10:46 PM | No Comments | Add Comment | Trackbacks (Suck)
Post contains 108 words, total size 1 kb.

Geek

It's How F'ing Much?!

So, if I'm going to be building a new PC, I'm going to be looking at the latest bits and pieces, right?

What I was thinking for the CPU was a low-power AMD chip, like the 9350e 2.0 GHz quad-core ($220) or the new model of the 5600+ 2.9GHz dual-core ($150).  Both are 65W chips - and 65W for a quad core is pretty nice.  The higher-clocked dual-core would be better for desktop stuff; the quad better for server stuff.  Shame they don't make a chip that lets you turn off two cores when you don't need them...

For the motherboard, either a Gigabyte GA-MA78GPM-DS2H ($160), a nice little all-in-one mATX board that's the successor to the successor to my current motherboards, or the GA-MA790GP-DS4H ($200), a full ATX board that has all the same features plus SLI support.  Both have built-in video, and include 128MB of dedicated DDR3 video ram.  Which is not critical now that I have a working video card, but is nice to have when I come to repurpose them in a year or two.

Throw in 8GB of memory ($160) and either a cheap 750GB drive ($125) or a shiny new 1.5TB drive ($290) and a case (which I have) and a power supply (which I also have, though I might buy a new one), and voila.

But hey, what about Intel's new Core i7?  Initial reports are that it's fast fast fast - about twice the speed of that AMD quad-core - and not too expensive, reportedly starting at US$284 in quantity.  It's only just been pre-announced, and I doubt anyone has it yet, but maybe they have pricing and an ETA...

Oh look!  It's in stock!  And it's only...  Uh, nearly $700?!  For the cheapest model, and $2400 for the high-end version.

Holy channel markup, batman.  I thought we'd left crap like that behind.

Motherboard?  One model in stock: The Gigabyte GA-EX58-EXTREME, a snip at just $600.

It takes 12GB of DDR3 memory, and to fill it up will cost, let's see, another $700.

So, $2000 for Intel CPU, motherboard and memory, vs. $470 to $580 for the AMD stuff.  The Core i7 is a lot faster, but it's not $1500 worth of faster, at least, not in my current financial situation.

Posted by: Pixy Misa at 02:26 AM | Comments (2) | Add Comment | Trackbacks (Suck)
Post contains 383 words, total size 2 kb.

Geek

I Hate It When That Happens, Part 46

So I was writing a chapter on using the CDN for The Minx Book, and I thought I'd get some accurate stats for my examples.

And I find that my ping time to the Sydney node has jumped from 14ms to 175ms, which is not materially better than the 215ms for the mee.nu server itself.

It's a good thing I'm not paying for this.  Oh wait, I am paying for this.

Grr! frown

Posted by: Pixy Misa at 12:11 AM | Comments (1) | Add Comment | Trackbacks (Suck)
Post contains 81 words, total size 1 kb.

Friday, November 07

Geek

Gonna Buy Me A Cuisinart

Well, no.  But I'm going to buy a new PC.

Install Vista, install everything else, copy all my data across, make sure it works, and then blow away the old one and put CentOS and OpenVZ on it.

Pity I didn't do it two months ago, before the prices all went up.

Posted by: Pixy Misa at 12:12 PM | No Comments | Add Comment | Trackbacks (Suck)
Post contains 57 words, total size 1 kb.

Geek

Anyone Want An Elf?

I seem to have one spare.

Update: Oops.  No, I have one elf and two collections of elf costumes.  Much less useful.

Posted by: Pixy Misa at 11:04 AM | Comments (5) | Add Comment | Trackbacks (Suck)
Post contains 26 words, total size 1 kb.

Thursday, November 06

Geek

ioDrive - Disappointment

Can't seem to break 65,000 iops for 4k random reads. sad

That's only the equivalent of 260 15,000rpm SCSI drives...

Heh.

4 threads (or aysnchronous I/O with a queue depth of 4) is enough to push the card pretty much flat out, which is great, because that means it's actually possible to get those benchmark results in real-world applications (if you don't simply end up CPU bound).

Too many threads, or too small or large a block size, tend to make it choke a bit, and it performs better on pure reads or writes than on a mixture, but the worst-case results still deliver in excess of 5000 iops.

It might not quite deliver on the marketing gloss on all points, at least not without more tweaking than I've had time to do so far.  I have been able to hit 700MB/sec on random reads (with a 16k block size), which is phenomenal.

Once I've collated the benchmark data, I'll produce some pretty graphs and a more detailed (and structured) review.  But right now, if you have a moderately large database and need better disk performance, the ioDrive is the way to go.

Posted by: Pixy Misa at 05:18 PM | No Comments | Add Comment | Trackbacks (Suck)
Post contains 196 words, total size 1 kb.

Wednesday, November 05

Geek

Meanwhile, Something That Matters

Testing the ioDrive again.  Getting 770MB/s.  On random reads.

Okay, sure, that's with a 128k block size and 4 threads, but it's still close to saturating the PCI-E slot.  At a nice steady 6000 iops.

I wrote a little Python script to generate 240 test cases with a combination of block sizes (512 bytes through 256kb), I/O activity (random or sequential, read, write or mixed), and threads(1, 2, 4 or 8), and I've set 'em off to run and log the results.

I planned to run the same set of tests on a normal drive for comparison.  Then I calculated how long that would take.  Then I quietly abandoned that plan...

Lessee now...  16k blocks, 1 thread gives me 14,877 iops; 2 threads gives me 24,908 iops; 4 threads gives 32,737 (and 524 MB/s throughput).  Shiny.

Posted by: Pixy Misa at 03:50 PM | No Comments | Add Comment | Trackbacks (Suck)
Post contains 140 words, total size 1 kb.

<< Page 2 of 2 >>
59kb generated in CPU 0.02, elapsed 0.5983 seconds.
53 queries taking 0.5854 seconds, 346 records returned.
Powered by Minx 1.1.6c-pink.
Using http / http://ai.mee.nu / 344