Dear Santa, thank you for the dolls and pencils and the fish. It's Easter now, so I hope I didn't wake you but... honest, it is an emergency. There's a crack in my wall. Aunt Sharon says it's just an ordinary crack, but I know its not cause at night there's voices so... please please can you send someone to fix it? Or a policeman, or...
Back in a moment.
Thank you Santa.
Back in a moment.
Thank you Santa.
Wednesday, October 31
Out Damn Singularity!
So my still-new Nexus 7 and my even newer New iPad are now both obsolete, replaced by the more capacityful New Nexus 7 and the fasteriffic New New iPad.
And here's the Nexus 10, just to complicate matters: One of the first Arm Cortex A15 devices to reach the market, 2GB RAM (twice as much as the Nexus 7), 16 or 32GB of flash (twice as much as the Old Nexus 7 but the same as the New Nexus 7), 2560x1600 display (vs. 1280x800), starting at $399. So cheaper, faster, and higher resolution than my Old New iPad. Which I only bought last month.
There's a problem, though, with the Nexus 7, Nexus 10, and the Nexus 4 - and shared with the iPad, iPad Mini, iPhone, and iPod Touch: Not one of these devices has any expansion capability. Between them, they have a total of zero micro-SD slots.
Given how fast these devices are coming out, how cheap SD cards are, how much Apple and Google charge for incremental upgrades, and how fast I filled up my Nexus 7, that's simply not good enough.
I was thinking of replacing my 16GB Nexus 7 with the 32GB model and giving the original to a family member, but now that Google has crippled the Nexus 10 and Nexus 4 the same way, I'm just getting ticked off.
The Asus Transformer Infinity is kind of expensive, but it has 64GB of flash built-in, a micro-SD slot in the tablet part, and a full-size SD slot in the keyboard dock. So I could have 256GB in total, today - and more later. Other expandable candidates include Samsung's Galaxy Note 10.1, Galaxy Note 2, and Galaxy S3. The Note 2 isn't available in Australia yet, but I can add the S3 to my sim-only mobile plan pretty cheaply.
If I could expect a new device to last more than a few months without becoming obsolete, I wouldn't mind so much. But since that's not the case, I do. Mind. So much.
Update: For crying out loud... I posted this yesterday. Since then, the Note 2 has been announced for Australia (November 14 release date) and now it looks like there's a Note 7 coming out. I'd much rather have a Note 7 than just a larger capacity Nexus 7, so I'll put that on hold. But if iiNet get the Note 2 (and they already have the Note and S3, so it's highly likely) then I'll get that as a phone, and there's a question of whether I really need a 5.5" phone and a high-end 7" tablet. Want, yes. Need... Probably yes. But then the Nexus 10 launches on November 13....
If Samsung can merge the Note 2 with the Nexus 7 for a Note 7 (preferably 1080P), and the Note 10.1 with the Nexus 10 for a Note 10.2, I'll happily buy all three sizes, because that would tick every box on my wish list. (32GB, SD card, stylus, brighter colours, higher resolution, and HDMI out.)
I actually have a couple of new wishlist items after using the Nexus 7 for longer: 2GB RAM (already implemented in the Note 2 and Nexus 10) and a Cortex A15 (found in the Nexus 10). The memory needs became apparent when I found that the simple puzzle game Cut the Rope was using 100MB of memory - even when I wasn't playing it. The CPU requirements are mostly from browsing complex web sites - Chrome on the Nexus 7 is maybe 5x slower than my desktop PC.
Posted by: Pixy Misa at
06:31 PM
| Comments (8)
| Add Comment
| Trackbacks (Suck)
Post contains 595 words, total size 4 kb.
Saturday, October 27
Jaded
These days I read High Scalability about how the world's largest websites manage their data, and I think, Oh, is that all?
These days I read High Scalability about how the world's largest websites manage their data, and I think, Oh, is that all?
28TB of MySQL? 100 billion rows? We* added that much when we put a new feature into the latest release of one of our products. We're adding storage in quarter-petabyte increments.
* At my day job.
Posted by: Pixy Misa at
06:22 AM
| Comments (4)
| Add Comment
| Trackbacks (Suck)
Post contains 59 words, total size 1 kb.
Saturday, October 13
Fuzzy Logic
The following fuzzies were served to you directly out of a zip file - testing a new feature.
Posted by: Pixy Misa at
11:58 PM
| Comments (3)
| Add Comment
| Trackbacks (Suck)
Post contains 20 words, total size 1 kb.
Thursday, October 11
VMWare Fixes Things (Somewhat)
I had an earlier rant about VMWare's pricing on Workstation 9: Not only was the pricing 25% higher than version 8, but they now had an Australian store that charged 40% more than the US store, for a price increase of 75%.
I had an earlier rant about VMWare's pricing on Workstation 9: Not only was the pricing 25% higher than version 8, but they now had an Australian store that charged 40% more than the US store, for a price increase of 75%.
The Australian store now only charges 4% more than the US store. Given that we have a national sales tax of 10% and the Aussie dollar is currently about 3 cents over parity, that's reasonable enough.
Posted by: Pixy Misa at
04:00 PM
| Comments (2)
| Add Comment
| Trackbacks (Suck)
Post contains 82 words, total size 1 kb.
Sunday, October 07
Now For The Fun Part!
Had some problems setting up Sakura, the new backup server. I could not get an ext4 filesystem to build under LVM on software RAID-5. It worked fine for the filesystems I set up at install time, but once the server was up, creating more pretty much made the server lock up.
Had some problems setting up Sakura, the new backup server. I could not get an ext4 filesystem to build under LVM on software RAID-5. It worked fine for the filesystems I set up at install time, but once the server was up, creating more pretty much made the server lock up.
Then I updated the kernel and suddenly it all worked. Hmm.
Now I just need to transfer about 15 million files across and we're good to go. Good to start getting ready to go, anyway. Slight problem in that this will use up most of our monthly bandwidth allocation.* But eh, we'll see how it goes.
* On the old servers. We have a ton of bandwidth on the new servers.
Posted by: Pixy Misa at
09:14 AM
| No Comments
| Add Comment
| Trackbacks (Suck)
Post contains 127 words, total size 1 kb.
Saturday, October 06
Zoomy Is As Zoomy Does
Huh. Well, the numbers don't lie. Probably.
Huh. Well, the numbers don't lie. Probably.
My new servers at ReliableSite in New Jersey - Aoi and Midori - use Xeon E3-1245 v2 chips; the ones at Incero in Dallas - Akane and Mikan - use Xeon E3-1270 v2s. There's only 100MHz between them - 3.4GHz and 3.5GHz respectively; the 1245 has integrated graphics while the 1270 does not, so the 1270 uses slightly less power but needs a video card or a motherboard with integrated graphics.
I have a Xeon E3-1230 as well, the previous generation at 3.2GHz.
The 1270 v2 runs my Python benchmark 15% faster than the 1230 v1, which is ~10% clock speed and 5% architecture. Fine.
The 1245 v2 runs the benchmark 25% faster than the 1230 v1. I checked several times, and it's consistent. Don't know why, but it's consistent. It looks like Incero and ReliableSite are using different motherboards, but that shouldn't make a difference to this benchmark.
It's also in line with the Passmark benchmark results that show the E3 v2 chips competing with the low-end Opteron 6200. They don't have the same memory capacity or bandwidth, and can't scale past a single socket, but they're certainly fast.
And we now have five of them. Well, I broke one, so four right now, but I'll have that fixed tomorrow.
Posted by: Pixy Misa at
02:16 AM
| No Comments
| Add Comment
| Trackbacks (Suck)
Post contains 223 words, total size 2 kb.
Friday, October 05
Unwanted Firsts
Today I installed a CentOS 6.3 server - in Dallas - from an ISO on my PC - in Sydney - using the IPMI virtual USB device.
Today I installed a CentOS 6.3 server - in Dallas - from an ISO on my PC - in Sydney - using the IPMI virtual USB device.
Okay, I took the easy route and did a net install from the mirror at Limestone Networks (also in Dallas), so it only took 40 minutes to get to the package selection menu and after that everything was nice and quick. But then I decided I wanted to rename my logical volumes, got it wrong, broke everything, had to reinstall, had Anaconda freak out and crash, had to reinstall again, had to go out to the shops while I was still waiting for the package selection menu to come up, had to reinstall again, and finally got things exactly the way I want them.
In short: It works, but it's not something you would ever want to do.
Posted by: Pixy Misa at
08:59 PM
| No Comments
| Add Comment
| Trackbacks (Suck)
Post contains 147 words, total size 1 kb.
Wednesday, October 03
Five Little Maids...
Akane and Mikan came online yesterday, but are offline again for some reconfigurationising.* They should be back soon.
Akane and Mikan came online yesterday, but are offline again for some reconfigurationising.* They should be back soon.
Aoi and Midori have been ordered and will be provisioned soon as well.
Sakura is still being set up. Kurumi is still around, doing nothing, but will probably be replaced with an equivalent config to Sakura (so twice its current disk and memory.)
Those six servers in all will cost me about as much as I currently pay for two, while delivering more of everything. I'll have 20TB of disk (all in RAID), 192GB of RAM, 1TB of SSD (all RAIDed), and 120TB of bandwidth.
Akane, Mikan, and Sakura will form one cluster in Dallas; Aoi, Midori, and Kurumi a second cluster in New Jersey.
The change in plans from earlier this year was sparked by plummeting SSD prices - 50% in just six months. I need a pair of SSDs in each server for performance and redundancy, and back in April that was expensive, pushing me towards a small number of large servers. Now I can get two servers, each with 32GB memory, two 2TB disks and two 256GB SSDs, for nearly the price of one server with 64GB of memory and that same storage configuration from before. So I get twice the storage and twice the bandwidth - and according to Passmark, twice the CPU performance** for about 10% more.
Update: Deployed! 2 x E3-1270 v2 in Dallas, 2 x E3-1245 vs in New Jersey.
* Either the technician wasn't paying close attention or their system is automated, because I ended up with my disks RAIDed together with my SSDs. Which works just fine, but isn't particularly useful. (Update: Apparently it's automated, and since I ticked the box for RAID-1, that's what the provisioning system did. They're manually reinstalling the servers now.)
** Passmark's results look a bit weird sometimes; they're averages of user-submitted results, and they change over time for reasons I can't explain. Anyway, I'm replacing the Opteron 6272 in New Jersey with a pair of Xeon 1245 V2s, and putting in another pair of 1270 (possibly V2s as well) in Dallas. Passmark put all those CPUs at about the same score, where my benchmarks suggest the 6272 should be at least 30% faster. We'll see when next one of the little maids comes online.
Posted by: Pixy Misa at
02:27 PM
| No Comments
| Add Comment
| Trackbacks (Suck)
Post contains 390 words, total size 3 kb.
Tuesday, October 02
Trinity Or Not Trinity?
AMD's new "Trinity" desktop chips are out, and there's been a bit of a flap over AMD's week-long "don't mention the CPU" embargo.
AMD's new "Trinity" desktop chips are out, and there's been a bit of a flap over AMD's week-long "don't mention the CPU" embargo.
Basically, the AMD chips clobber Intel on integrated graphics performance once again, but lag behind on pure CPU tasks. SemiAccurate compares the top-of-the-line Trinity to the top-of-the-line Ivy Bridge, and shows that Trinity is only 60% as fast on CPU-intensive workloads.
However, it's also only 40% as expensive as the top-of-the-line Intel chip, and as SemiAccurate note, you can't see the difference in CPU speed in day-to-day tasks, but you can see the difference in GPU speed. Games that are unplayable on Intel's integrated graphics run relatively smoothly on AMD's chips. Of course, an Intel chip plus a dedicated graphics card will win over the AMD chip alone every time, but that costs more again.
Another thing to consider is that if the game you're playing is a console port, AMD's integrated graphics now clearly outclass current-generation consoles. The higher-end Trinity parts deliver 600 GFLOPS vs. 240 GFLOPS for the Xbox 360's GPU.
Anandtech examines the chips in two parts: gaming and graphics performance and CPU performance.
Against the competitively-priced Intel Core i3, the CPU results are mixed, though the Intel chip does show better single-threaded performance. The GPU results are entirely one-sided, though, with the AMD chips showing two, three, four, and in the case of Civilization V* a humiliating seven times the speed of Intel's integrated graphics.
They also test power consumption of the AMD vs. the Intel chips running the game Metro 2033, and AMD fares poorly there, using twice as much power overall. But you have to consider that it's also running the game twice as fast...
Looking at the SemiAccurate article I can understand AMD's feelings here - you don't want the first stories about your new chip to say that it's 40% slower than the competition - even though the competition is more than twice the price. I'm less convinced that putting an embargo on reality is the solution to that problem.
In all, though, this is the chip I'd choose for a general-purpose desktop or HTPC build, and probably for a home server as well.** Specifically the A10 5700, a 65W part that's only slightly slower than the 100W A10 5800K. For my higher-end machines, I'm still awaiting the FX 8350, which is due... Soonish.
* I have a free copy of this on Steam, if any of my regular readers are interested.
** Servers don't need the GPU power - not unless you're running floating point apps that can take advantage of it as a GPGPU. The reason AMD is better for small servers is all down to Intel's marketing.
Intel have a nasty practice of market fragmentation; their slower and cheaper CPUs aren't just slower and cheaper, they are actually missing features found on their higher-end chips, features that are relevant to servers. For example, Trinity is up to 6x as fast as a Core i3 for encryption, because AMD support hardware-accelerated encryption on all their chips, while Intel only enable it on select models. The Core i3 also lacks VT-d, which you really want if you're running virtual servers, and Trusted Execution ("TXT"), which you probably don't want at all. And it's clock-locked, so no overclocking for you! Though Trinity isn't a great overclocker anyway. Unless you douse it in liquid nitrogen; then you can hit 8GHz.
Posted by: Pixy Misa at
09:42 PM
| Comments (23)
| Add Comment
| Trackbacks (Suck)
Post contains 574 words, total size 4 kb.
Monday, October 01
This Is All Your Fault, Google
I've been buying five new Android apps a day recently.
I've been buying five new Android apps a day recently.
Thanks to Google, that's been costing me $1.25. They're having a bit of a sale - to celebrate 25 billion downloads, they're offering a new selection of top apps each day at just 25¢ each.
Today's purchases included the well-regarded adventure game HORN. 25¢. 1.7GB download. My Nexus 7 only has 16GB of flash, total, and about 13GB available after the OS and reserved space. A 25¢ app just ate 13% of the space on my $250 device. (Actually, it's US$250, but A$300.)
I also bought The Dark Knight Rises Batman game. I wasn't going to at first, then I realised that at 25¢, the time I'd spent thinking about it had already cost me more than the game itself. That's another 1.8GB. After HORN, I don't think I have that much space left.
64GB model now, please. Or better, 128GB. I'd even consider buying a 256GB model if the price differential was competitive to SSDs - i.e. below $1/GB.
Also, microSD is not a luxury in these devices, it's essential.
Posted by: Pixy Misa at
04:10 PM
| No Comments
| Add Comment
| Trackbacks (Suck)
Post contains 187 words, total size 2 kb.
<< Page 1 of 1 >>
90kb generated in CPU 0.0587, elapsed 0.2149 seconds.
55 queries taking 0.1975 seconds, 392 records returned.
Powered by Minx 1.1.6c-pink.
55 queries taking 0.1975 seconds, 392 records returned.
Powered by Minx 1.1.6c-pink.
