I code on a PS3 running Linux. There are 7 cores available to Linux. 1 PPU + 6 SPU cores.
Ok the PPU is not as powerful as an SPU, it's a basic in-order dual-threaded PowerPC core with the AltiVec instruction set, but you shouldn't ignore it.
I would add that virtually every email service provide already reads your emails... how does Balmer think anti-spam technogologies work ? Duh ! (Notable exception: greylisting.)
You thought wrong. The laptop is now known as the XO-1 laptop (they have not been able to maintain the price within the original estimation, it is today priced at $188).
As a side note, IMHO the software development and integration efforts that are happening on the OLPC project are fantastic. All the companies involved in this project are providing their best engineers: Marvell (who made the wireless chip) have their guys developing the firmware often directly according to the feedback they get from the kernel developers, Red Hat is providing plenty of sw engineers (including Marcelo Tosati, who was the 2.4 kernel maintainer!), AMD and Quanta are working on the hardware platform (recently they made efforts to track the power consumption of every single chip in the laptop), etc. This is just incredible how fast the teams are able to progress in such a cooperative environment. This is a sharp contrast with what happens too often in the ordinary Linux world where cooperation is sometimes difficult or inexistent (e.g. kernel developers unable to obtain hardware specs, or hardware vendors attempting to provide some crappy binary drivers without involving the kernel community, etc).
I certainly expect a very high quality product to come out of this project, both on the hardware level and software level. Every single piece of chip or software has been optimized and fine-tuned to make the whole platform work as best as it can. This is going to be one of the best Linux laptop ever made. Just read their weekly updated news page to get a brief understanding of the technical achievements made possible in such an ultra-cooperative environment: http://wiki.laptop.org/go/News
...I require network traffic to use secure protocols (SSL/TLS, etc) on the internal networks I administer, even if they are protected from external attackers by a firewall. Use POP3S/IMAPS to prevent the employees from accessing others' mailboxes. Run your intranet website on HTTPS. Use LDAPS. Force CIFS connections to be signed and encrypted and to use LMv2/NTLMv2.
This chart depicts how Sun used to be. But since Schwartz became Sun's CEO in April 2006, the company has been actually very focused on its
goals (opensourcing its OS, CPUs, languages, increasing R&D, etc).
Of particular interest is how Schwarts made the point in one of its recent blog entries, that his company is the perfect example of how you can make money from opensource products (since he became CEO, Sun has become profitable):
We did this while driving significant product transitions, going after new markets and product areas, and best of all, while aggressively moving the whole company to open source software (leading me to hope we can officially put to rest the question, "how will you make money?").
Let me say "wow", what an insightful advice ! None of our top-notch engineers had thought about that before. Would you consider joining one of our engineering teams ? We feel you could be a precious asset to the company.
Thanks,
Intel.
PS: Please don't tell AMD about this extraordinary good idea.
Do you refer to Vista by its codename "Longhorn" ? No ! So why do you do it with Ubuntu ?
Why do you insist in calling it "Feisty Fawn", and then you complain people laugh at you, when you could just say "Ubuntu 7.04" ?
I am truly perplexed.
No buffer size tweaking of any sort can improve the situation. This is just fundamental math: if you receive 1500-byte packets at a throughput of 800 Mbps, your NIC needs to process 67000 packets/sec. If you tweak the NIC settings to generate no more than 5000 interrupts, then each interrupt will deliver on average 13 packets. Which means that before executing the NIC driver IRQ handler, up to 13 packets will have accumulated in the NIC hardware buffers. So the oldest packet in the hw buffer will have spent 1/5000 = 0.2 ms there before the kernel will even receive it. (The latest receive packet will have spent a negligible amount of time in this buffer, because the IRQ is generated right after receiving it). So on average a packet will stay 0.1 ms in this hardware buffer before being processed by the kernel.
How can changing a buffer size somewhere improve the situation ? It cannot. The only way to reduce latency caused by an interrupt rate moderation mechanism is to tweak it to generate interrupts slightly more frequently. Which is why, for example, the e1000 Linux driver chooses a good compromise of 9000 int/sec by default.
Oh I have seen crappy Nvidia GbE controllers generate more than 50,000 int/sec at around 700 Mbps...
Of course better NICs behave much better. For example the Linux e1000 driver for Intel GbE controllers configures their interrupt moderation mechanism to generate no more than about 9,000 int/sec by default when saturating a GbE link. This is the only reason why I mentionned 10,000 as the lower boundary. But yes, as you experienced it, it is of course possible to reduce this to 5,000 int/sec and probably even lower values. You don't necessarily want to do it though because 5,000 int/sec and lower means it would add, on average, 1/5000/2 = 0.1 ms to the network latency (which is about 0.1 ms on a typical GbE LAN, so it would double it).
All the open source media players I am aware of implement the pause feature the same way: by feeding "silence" to the sound card.
So by pausing an MP3 you save less than 1% of CPU time decoding your MP3 stream (negligible), but the whole userspace audio subsystem
is still sending 48,000 x 2 (stereo) 16-bit samples per second to the kernel...
So I would say it actually reinforces my theory of the audio drivers being in userspace causing this pb.
I am surprised no slashdotter mentionned this already... But could it be caused by the fact that, in Vista, the audio drivers are implemented in userspace ? My guess is that an actively used audio driver in userspace causes roughly 5,000 to 10,000 extra context switches per second. I didn't RTFA but this kind of CPU overhead would definitely be big enough to cause a visible reduction in network throughput when trying to max out a GbE link... Either because of the CPU time spent dealing with the context switches, or the extra latency it can introduce if some locks have to be held too long by the Vista kernels on some data structures concurrently used by the audio and network layer. Keep in mind that GbE network cards generate roughly 10,000 to 50,000 interrupts/sec when transferring at speeds approaching 1 Gbit/s, so a low latency in processing these IRQ is also critical.
Contrary to common belief, power-of-10 prefixes (as used by disk manufacturers) are much more commonly used than power-of-2 prefixes in the IT world. People claiming the contrary are wrong. Here are a few examples:
A 128 kbit/s audio stream is 128 * 10^3 bit/s (power of 10)
A 100 Mbit/s ethernet card is 100 * 10^6 bit/s (power of 10)
A 480 Mbit/s USB2 link is 480 * 10^6 bit/s (power of 10)
A 500 GByte disk is 500 * 10^9 bytes (power of 10)
A 56 kbaud modem is 56 * 10^3 baud/s (power of 10)
A 1.5 GHz processor is 1.5 * 10^9 Hz (power of 10)
A 6 Mbit/s DSL line is 6 * 10^6 bit/s (power of 10)
A 650 MByte CD is 650 * 10^6 bytes (power of 10)
It is a total mystery to me why people think that power-of-2 prefixes should be the norm, when the only few places where they are used are to refer to the size of files and RAM sticks.
Actually ZFS is very attractive even on single-disk systems. Some ZFS
developers are using it on their laptop because:
ZFS keeps redundant copies of metadata (directory entries, inodes, etc). Which
means, for example, that if only 1 block causes a corruption of the/home
inode, a traditional FS would typically crash and prevent you from accessing
anything under that dir (even if that data is healthy), whereas ZFS will use
the redundant copy, self-heal itself by rewriting the 1st copy, and let
you access the data under/home.
You can also do zfs set copies=2 pool/fs to make use of the above
feature for regular data. Of course this double the required disk space, and is
by no mean a raid1 replacement, but you may use that feature to improve data
reliability even with a single disk.
And don't forget all the other ZFS features: detection and correction of silent
data corruption (especially important if your data go through a noisy USB
cable), expandability of the pool (add a 2nd drive to your existing single USB
drive to make use of raid1: "zpool attach pool disk1 disk2"), no more messing
with partitions (ability to dynamically resize filesystems...), simpler to
administer, etc.
Heck, there are lots of reasons for using ZFS on
single-disk systems !
#!/usr/bin/perl -w use strict; print "You website is slow because: your (average) webmaster/sysadmin/architect cannot ". "tell the difference between www.thedailywtf.com and good code\n";
AMD have indicated they may do MCM in the future (Multi-Chip Module), like Intel. But since they are releasing a true quad-core CPU before Intel, it is going to give them an advantage: to make an 8-core CPU, they will just need 2 quad-core chips, whereas Intel will need 4 dual-core chips.
I wonder how significant this technical advantage really is on various levels: performance, power consumption, reliability, yield, simpler to manufacture, cost, etc. Could this also mean AMD will be first to market with 8-core CPUs ?
Assuming a non-RAID 3x-replication tech solution (what Google do in their datacenters), using 500-GB disks (best $/GB ratio), they would need about 16 thousands disks:
Now for various reasons (vendors influence, inexperienced consultants, my experience in the IT world in general, etc), I have a feeling they are going to end up with a solution unnecessarily complex, much more expensive, and hard to maintain and expand... Damn, I would love to be this project leader !
Vulnerability Info Exchange is Good
on
An eBay For Hackers
·
· Score: 2, Interesting
Selling information about security vulnerabilities may be considered unethical by some, but it is perfectly legal in almost all countries (notable exception: France).
Don't forget that a vuln is just a bug, they are selling information about how to trigger a bug. Why would that be illegal ? If a buyer exploit the bug for nefarious purposes, then the buyer is doing something illegal, not the seller. There are plenty of legitimate cases where a market for selling vulnerabilities is a good thing:
The developer of a vulnerable application may want to buy vulnerabilities found in his application. Financial reward is an incentive for security researchers to find more vulnerabilities in an application when they know they would get paid for it. Additionally, over time, the security of the application increases.
Penetration testing companies can increase their chance of a successful pentest with access to new and original vulnerabilities. This prompts the client (pentest target) to secure his IT infrastructure, by rearchitecting it, or implementing new security mechanisms.
Vulnerability assessment tools developers can gain an edge over their competitors by buying info about vulns. A legitimate security vuln market moves the VA market forward, and in the end increases the rate of discovery, and therefore the rate at which security vulns are fixed.
The biggest argument is perhaps this one: a vulnerability sold to a legitimate buyer is often one less sold to a criminal.
Slashdot Mirror
I code on a PS3 running Linux. There are 7 cores available to Linux. 1 PPU + 6 SPU cores.
Ok the PPU is not as powerful as an SPU, it's a basic in-order dual-threaded PowerPC core with the AltiVec instruction set, but you shouldn't ignore it.
I would add that virtually every email service provide already reads your emails... how does Balmer think anti-spam technogologies work ? Duh ! (Notable exception: greylisting.)
You thought wrong. The laptop is now known as the XO-1 laptop (they have not been able to maintain the price within the original estimation, it is today priced at $188).
As a side note, IMHO the software development and integration efforts that are happening on the OLPC project are fantastic. All the companies involved in this project are providing their best engineers: Marvell (who made the wireless chip) have their guys developing the firmware often directly according to the feedback they get from the kernel developers, Red Hat is providing plenty of sw engineers (including Marcelo Tosati, who was the 2.4 kernel maintainer!), AMD and Quanta are working on the hardware platform (recently they made efforts to track the power consumption of every single chip in the laptop), etc. This is just incredible how fast the teams are able to progress in such a cooperative environment. This is a sharp contrast with what happens too often in the ordinary Linux world where cooperation is sometimes difficult or inexistent (e.g. kernel developers unable to obtain hardware specs, or hardware vendors attempting to provide some crappy binary drivers without involving the kernel community, etc).
I certainly expect a very high quality product to come out of this project, both on the hardware level and software level. Every single piece of chip or software has been optimized and fine-tuned to make the whole platform work as best as it can. This is going to be one of the best Linux laptop ever made. Just read their weekly updated news page to get a brief understanding of the technical achievements made possible in such an ultra-cooperative environment: http://wiki.laptop.org/go/News
...I require network traffic to use secure protocols (SSL/TLS, etc) on the internal networks I administer, even if they are protected from external attackers by a firewall. Use POP3S/IMAPS to prevent the employees from accessing others' mailboxes. Run your intranet website on HTTPS. Use LDAPS. Force CIFS connections to be signed and encrypted and to use LMv2/NTLMv2.
The wife: Don't you think I am gaining weight ?
Me: No honey, it's just the kilogram that is getting lighter.
She is indeed a Sony fembot. Ahh that could explain a lot of things...
Every time she tries to use a laptop, it melts because... she is so hot.
Mr. rossdee,
Let me say "wow", what an insightful advice ! None of our top-notch engineers had
thought about that before. Would you consider joining one of our engineering teams ?
We feel you could be a precious asset to the company.
Thanks,
Intel.
PS: Please don't tell AMD about this extraordinary good idea.
Which makes Steve a great artist !
Do you refer to Vista by its codename "Longhorn" ? No ! So why do you do it with Ubuntu ? Why do you insist in calling it "Feisty Fawn", and then you complain people laugh at you, when you could just say "Ubuntu 7.04" ? I am truly perplexed.
No buffer size tweaking of any sort can improve the situation. This is just fundamental math: if you receive 1500-byte packets at a throughput of 800 Mbps, your NIC needs to process 67000 packets/sec. If you tweak the NIC settings to generate no more than 5000 interrupts, then each interrupt will deliver on average 13 packets. Which means that before executing the NIC driver IRQ handler, up to 13 packets will have accumulated in the NIC hardware buffers. So the oldest packet in the hw buffer will have spent 1/5000 = 0.2 ms there before the kernel will even receive it. (The latest receive packet will have spent a negligible amount of time in this buffer, because the IRQ is generated right after receiving it). So on average a packet will stay 0.1 ms in this hardware buffer before being processed by the kernel.
How can changing a buffer size somewhere improve the situation ? It cannot. The only way to reduce latency caused by an interrupt rate moderation mechanism is to tweak it to generate interrupts slightly more frequently. Which is why, for example, the e1000 Linux driver chooses a good compromise of 9000 int/sec by default.
Oh I have seen crappy Nvidia GbE controllers generate more than 50,000 int/sec at around 700 Mbps...
Of course better NICs behave much better. For example the Linux e1000 driver for Intel GbE controllers configures their interrupt moderation mechanism to generate no more than about 9,000 int/sec by default when saturating a GbE link. This is the only reason why I mentionned 10,000 as the lower boundary. But yes, as you experienced it, it is of course possible to reduce this to 5,000 int/sec and probably even lower values. You don't necessarily want to do it though because 5,000 int/sec and lower means it would add, on average, 1/5000/2 = 0.1 ms to the network latency (which is about 0.1 ms on a typical GbE LAN, so it would double it).
All the open source media players I am aware of implement the pause feature the same way: by feeding "silence" to the sound card. So by pausing an MP3 you save less than 1% of CPU time decoding your MP3 stream (negligible), but the whole userspace audio subsystem is still sending 48,000 x 2 (stereo) 16-bit samples per second to the kernel...
So I would say it actually reinforces my theory of the audio drivers being in userspace causing this pb.
I am surprised no slashdotter mentionned this already... But could it be caused by the fact that, in Vista, the audio drivers are implemented in userspace ? My guess is that an actively used audio driver in userspace causes roughly 5,000 to 10,000 extra context switches per second. I didn't RTFA but this kind of CPU overhead would definitely be big enough to cause a visible reduction in network throughput when trying to max out a GbE link... Either because of the CPU time spent dealing with the context switches, or the extra latency it can introduce if some locks have to be held too long by the Vista kernels on some data structures concurrently used by the audio and network layer. Keep in mind that GbE network cards generate roughly 10,000 to 50,000 interrupts/sec when transferring at speeds approaching 1 Gbit/s, so a low latency in processing these IRQ is also critical.
You are right. I should have said 56 kbps. V.92 modems run at 8000 baud.
Power-of-10 prefixes are the norm in IT
-
A 128 kbit/s audio stream is 128 * 10^3 bit/s (power of 10)
-
A 100 Mbit/s ethernet card is 100 * 10^6 bit/s (power of 10)
-
A 480 Mbit/s USB2 link is 480 * 10^6 bit/s (power of 10)
-
A 500 GByte disk is 500 * 10^9 bytes (power of 10)
-
A 56 kbaud modem is 56 * 10^3 baud/s (power of 10)
-
A 1.5 GHz processor is 1.5 * 10^9 Hz (power of 10)
-
A 6 Mbit/s DSL line is 6 * 10^6 bit/s (power of 10)
-
A 650 MByte CD is 650 * 10^6 bytes (power of 10)
It is a total mystery to me why people think that power-of-2 prefixes should be the norm, when the only few places where they are used are to refer to the size of files and RAM sticks.Spread the truth. Mod me informative
Actually ZFS is very attractive even on single-disk systems. Some ZFS developers are using it on their laptop because:
Heck, there are lots of reasons for using ZFS on single-disk systems !
Of course it will.
Nicholas Negroponte: "It's an education project, not a laptop project."
AMD have indicated they may do MCM in the future (Multi-Chip Module), like Intel. But since they are releasing a true quad-core CPU before Intel, it is going to give them an advantage: to make an 8-core CPU, they will just need 2 quad-core chips, whereas Intel will need 4 dual-core chips.
I wonder how significant this technical advantage really is on various levels: performance, power consumption, reliability, yield, simpler to manufacture, cost, etc. Could this also mean AMD will be first to market with 8-core CPUs ?
Assuming a non-RAID 3x-replication tech solution (what Google do in their datacenters), using 500-GB disks (best $/GB ratio), they would need about 16 thousands disks:
Which would cost about $1.8M (disks alone):
15552 (disk) * 110 ($/disk) = $1710720
Packed in high-density chassis (48 disks in 4U, or 12 disks per rack unit), they could store this amount of data in about 30 racks:
15552 (disk) / 12 (disk/rack unit) / 42 (rack unit/rack) = 30.9 racks
Now for various reasons (vendors influence, inexperienced consultants, my experience in the IT world in general, etc), I have a feeling they are going to end up with a solution unnecessarily complex, much more expensive, and hard to maintain and expand... Damn, I would love to be this project leader !
Selling information about security vulnerabilities may be considered unethical by some, but it is perfectly legal in almost all countries (notable exception: France). Don't forget that a vuln is just a bug, they are selling information about how to trigger a bug. Why would that be illegal ? If a buyer exploit the bug for nefarious purposes, then the buyer is doing something illegal, not the seller. There are plenty of legitimate cases where a market for selling vulnerabilities is a good thing: