You get to a point "most of the way through" the game and it becomes nonlinear until you decide to continue the plot.
That said, its always like that -- you can deviate and go off course at any point, or backtrack, or whatever -- but the game designers have a specific plot they want you to follow. Ever notice how you can't kill off a team member permanently?
Slashdot has a lot more to do with the comments on the stories than the stories themselves.
I rarely read the stories before reading others' perceptions unless I'm a personal expert on the issue in question.
I'd rather read what Bruce Perens thinks of a programming book that gets written about, or what John Carmack has to say about the latest 3D thingamajig than what Cowboy Neal thinks.
I'm glad Slashdot exists for the sake of the community; not because the articles are always well chosen or edited.
The refund doesn't help -- I still want the source.
Re:Transport latency and TCP
on
Replacing TCP?
·
· Score: 1
My comment was that instead of "learning" the appropriate window size on each connection, remember an appropriate starting window size for the next time.
On long-running servers, I'm betting this would make a noted difference.
Re:Transport latency and TCP
on
Replacing TCP?
·
· Score: 1
I will comment before reading your paper... but wouldn't a kernel remembering appropriate window sizes for given end-points decrease latency as well? Nobody seems to be talking about TCP window sizes at all here for that matter, and they're of vital importance to both throughput and latency.
Given two endpoints with high network latency (1s rtt), high bandwidth (1Gbps) and high reliability (no packet loss), the TCP window size set too low will cause huge loss in potential bandwidth use since the sender will be waiting for ACKs on data that is being sent successfully before sending more data instead of filling the pipe appopriately.
If the sender "learned" appropriate window sizes for given receivers and reused them on future connections, wouldn't the streams work better the first time more often?
I've seen some nice algorithms for improved versions of RED that try to optimize window sizes more rapidly as well -- (thanks Mike) -- and replacing packet drop with ECN can eliminate lost packets almost entirely.
... not at all like how they define a TRY macro to use instead of try for C++ (although they both work). Or how they define new classes to do things the standard template library does very well. Not at all:)
I'm sure you actually realize that every industry puts their own meanings on terms like MTBF and you actually read my links to how hard drives are actually rated.
I miss the private BBS accounts I had that required actually calling people and begging for an account. Or accounts that required regular contributions to keep open.
Remember U/D ratios? Can you imagine if fileplanet had a U/D ratio? I'm sure that's what caused a lot of the content to even exist -- people needed to create stuff to upload to be able to download the other things they wanted. Everyone had to contribute something.
You obviously never sat in front of your screen download a one meg file at 232 bytes per second and watching for Ymodem-G to pull off a 233 or 234 every now and then.
I spent over $300/mo between the ages of 12 and 15 connecting to BBSs worldwide and hosting a northern-Ontario FidoNET (and BoM and ACiD) node. It was a great experience and I enjoyed it tremendously. Its helped me become the network designer I am today.
As an aside, I use Multitech modems to this day for my clients for three reasons: they're technologically superior* to any other product I've found so far, they're rugged and have lots of hardware options (for rack-mounting, etc.) and most importantly, their support is outstanding.
I ran into some trouble with an MT5600BA not working with a remote fax machine with HylaFAX and the HylaFAX support people (also excellent) connected me with Multitech who then sent me a custom firmware update to fix the problem. Thanks Multitech -- worth the extra $$$ every time.
* Aside from other features you'll find in their manuals online, their modems support retraining to higher speeds as well as lower. When the line noise on a long-term connection drops the speed down to 1200 baud but then goes away, most modems will stay at 1200 baud until disconnected and reconnected. A Multitech modem will retrain up to faster speeds as it detects better SnR conditions.
I replied to (presumably) your anonymous post as well, but just so you can learn something *specifically* about MTBF ratings in the drive industry, please read:
Top 10 Causes of Data Loss (from the June 2003 issue of ComputerWorld Magazine) Rank Cause Weight* 1 Mechanical hard-drive failure 24.7 2 Data structure corruption 16.5 3 Accidental or intentional data deletion 12.4 4 RAID server failure 9.9 5 Backup tape malfunction 8.2 6 Physical tape damage 6.3 7 Accidental overwrites 6.2 8 Software corruption 5.5 9 Viruses 4.9 10 Natural disasters 4.4 (*Weights computed using Zipf's Law for ranked items--the probability of many ranked events is inversely proportional to their rank.)
Excuse me, but how do you intend to replace the platter that dies? Or do you want to toss the whole disk? If so, how are you going to get a new mirror copy?
RAID-1 and RAID-5 work because you can remove the failed component and replace it offline or online and rebuild the data and parity as applies.
Any script kiddie can download enterprise class source code:)
Priceless is running it on your home computer, finding a bug, fixing it, and having Linus Torvalds tell you your patch is going in and millions of other people will see it in an hour.
I don't mind the rating being relevant to parents' views at all -- the parents with the girl who's putting out for the football team is probably oblivious to that fact.
Slashdot Mirror
And why pray-tell, would one believe that?
Do you honestly believe that male-female only marriages are a north-american or western or christian-only tradition?
Get an education.
You get to a point "most of the way through" the game and it becomes nonlinear until you decide to continue the plot.
That said, its always like that -- you can deviate and go off course at any point, or backtrack, or whatever -- but the game designers have a specific plot they want you to follow. Ever notice how you can't kill off a team member permanently?
Slashdot has a lot more to do with the comments on the stories than the stories themselves.
I rarely read the stories before reading others' perceptions unless I'm a personal expert on the issue in question.
I'd rather read what Bruce Perens thinks of a programming book that gets written about, or what John Carmack has to say about the latest 3D thingamajig than what Cowboy Neal thinks.
I'm glad Slashdot exists for the sake of the community; not because the articles are always well chosen or edited.
The refund doesn't help -- I still want the source.
My comment was that instead of "learning" the appropriate window size on each connection, remember an appropriate starting window size for the next time.
On long-running servers, I'm betting this would make a noted difference.
I will comment before reading your paper ... but wouldn't a kernel remembering appropriate window sizes for given end-points decrease latency as well? Nobody seems to be talking about TCP window sizes at all here for that matter, and they're of vital importance to both throughput and latency.
Given two endpoints with high network latency (1s rtt), high bandwidth (1Gbps) and high reliability (no packet loss), the TCP window size set too low will cause huge loss in potential bandwidth use since the sender will be waiting for ACKs on data that is being sent successfully before sending more data instead of filling the pipe appopriately.
If the sender "learned" appropriate window sizes for given receivers and reused them on future connections, wouldn't the streams work better the first time more often?
I've seen some nice algorithms for improved versions of RED that try to optimize window sizes more rapidly as well -- (thanks Mike) -- and replacing packet drop with ECN can eliminate lost packets almost entirely.
The key is to market them to the kids to mow the lawn for them ;-) ... and to make it look like the kid's still the one doing the mowing.
I sure take X forwarding for granted.
I ssh into a server, run a GUI app, ssh into another server, run a different GUI app.
Actually, I double-click icons that do "ssh servername appname" and my ssh-agent handles the logins for me.
Yay X.
... not at all like how they define a TRY macro to use instead of try for C++ (although they both work). Or how they define new classes to do things the standard template library does very well. Not at all :)
I'm sure you actually realize that every industry puts their own meanings on terms like MTBF and you actually read my links to how hard drives are actually rated.
I miss the private BBS accounts I had that required actually calling people and begging for an account. Or accounts that required regular contributions to keep open.
Remember U/D ratios? Can you imagine if fileplanet had a U/D ratio? I'm sure that's what caused a lot of the content to even exist -- people needed to create stuff to upload to be able to download the other things they wanted. Everyone had to contribute something.
You obviously never sat in front of your screen download a one meg file at 232 bytes per second and watching for Ymodem-G to pull off a 233 or 234 every now and then.
I spent over $300/mo between the ages of 12 and 15 connecting to BBSs worldwide and hosting a northern-Ontario FidoNET (and BoM and ACiD) node. It was a great experience and I enjoyed it tremendously. Its helped me become the network designer I am today.
As an aside, I use Multitech modems to this day for my clients for three reasons: they're technologically superior* to any other product I've found so far, they're rugged and have lots of hardware options (for rack-mounting, etc.) and most importantly, their support is outstanding.
I ran into some trouble with an MT5600BA not working with a remote fax machine with HylaFAX and the HylaFAX support people (also excellent) connected me with Multitech who then sent me a custom firmware update to fix the problem. Thanks Multitech -- worth the extra $$$ every time.
* Aside from other features you'll find in their manuals online, their modems support retraining to higher speeds as well as lower. When the line noise on a long-term connection drops the speed down to 1200 baud but then goes away, most modems will stay at 1200 baud until disconnected and reconnected. A Multitech modem will retrain up to faster speeds as it detects better SnR conditions.
I replied to (presumably) your anonymous post as well, but just so you can learn something *specifically* about MTBF ratings in the drive industry, please read:
ASSUMPTIONS FOR RELIABILITY MODELS, part of Latent Sector Faults and Reliability of Disk Arrays, a dissertation by HANNU H. KARI of the Helsinki University of Technology.
Note for everyone though: "the actual reliability may vary dramatically and, for some units, MTBF can be only a fraction of the average MTBF."
You've got some problems reading numbers. MTBF numbers are how often you can expect a bit-level defect on a drive.
The *first* failure is to be expected much longer from now than that.
Yes, that means these drives are predicted to *never* fail at that level. Go read your Seagage or Fujitsu specs again.
I have several Fujitsu 10kRPM drives from over 5 years ago with 1Mhr MTBF ratings. They still have 0 grown defects.
Excuse me, but how do you intend to replace the platter that dies? Or do you want to toss the whole disk? If so, how are you going to get a new mirror copy?
RAID-1 and RAID-5 work because you can remove the failed component and replace it offline or online and rebuild the data and parity as applies.
You meant to say propaganda :)
I mentionned both on purpose ;-) Just in case someone didn't like my blue drill and prefered their yellow one.
And here I thought my Dewalt and Makita tools were better than Craftsman.
Any script kiddie can download enterprise class source code :)
Priceless is running it on your home computer, finding a bug, fixing it, and having Linus Torvalds tell you your patch is going in and millions of other people will see it in an hour.
I don't mind the rating being relevant to parents' views at all -- the parents with the girl who's putting out for the football team is probably oblivious to that fact.
Part 2: smarten up parents. Yeah right.
I actually prefer Windows 98 Second Edition myself, if I have to pick a version at all.
I run Linux exclusively at home.
Read some surveys -- those under 17 are most likely to have *participated* in oral sex, but not in shootings.
People under 17 are more likely to have had oral sex than vaginal intercourse. CNN even covered this last year.
Consider that if we won't show sexual parts of the body on TV, we probably shouldn't even show guns.
Would you rather let your kid see a girl without her shirt on, or holding a gun?
Which would you prefer in a movie?
And then we're taught to think those darn arabs are the violent ones.