Network stack developer Andre Opperman appeals to the FreeBSD community to support three months of full time work. The primary focus will be to optimize the IP network stack and estimates a potential perfomance increase of 40-50%.
Network stack developer Andre Opperman appeals to the FreeBSD community to support three months of full time work. The primary focus will be to optimize the IP network stack and estimates a potential perfomance increase of 40-50%.
I thought that premature optimization was the root of all evil. Although, maybe he’s not being premature?
> esstimates a potential perfomance increase of 40-50%.
The FreeBSD TCP/IP stack is known to perform very well and is among the best ones, if not the best one. How could they announce such perfomance increase ?
>> estimates a potential perfomance increase of 40-50%.
>
> The FreeBSD TCP/IP stack is known to perform very well
> and is among the best ones, if not the best one. How
> could they announce such perfomance increase ?
You obviously haven’t tested 5.x or 6.x. A 40-50% increase would *still* be slower than 4.x was. 4.x was indeed one of the best, if not the best. They lost a lot of ground there because of the introduction of SMPng, which provides performance gains under some work loads, but big losses under others and network throughput is one that suffered big losses.
this small bit of the much more extensive testing I’ve done clearly shows this. (6.x, last I tested a couple months back, was identical to 5.x and I didn’t include it in these particular results)
http://chrisbuechler.com/4vs5/
Try a recent DragonFly 1.3.x TCP/IP stack. Lots of changes and optimisations.
Will this ever be able to be used on something other than Mac/Windows?
8000 Swiss franks a month? he lives well over there! Switzerland is expensive but not that expensive. My family that lives there survives on much less. I truly dowbt he will optimize the code to produce 40-50% performance increase. Who is going to do the bug finding/testing?
Indeed. Switzerland is NOT twice as expensive to live in as Holland. Is he supporting a big family or something? Or does he require a high-paid programmers salary?
I’d not donate on the basis that he’s overcharging.
However he’s already got about a quarter in 8 days time. Wonder what happens if he doesn’t make true to his estimates. (Although one’d expect him to be relatively accurate if he already has 60-80% done of some parts).
why not? maybe he DESERVES THAT MUCH.
if he is worth it, then he is worth it.
That’s obviously not the salary he can take out of his company per month…
After loads of taxes, you get much less than that into your own pocket.
If it was in sweden, around 54% of the 5200Euro would disappear in taxes… Not all of it being income tax.
I don’t know what taxes apply for this guy, but don’t think he can actually take out as much as he asks for in pure salary.
He should explain these parts so more people would understand it.
8000 Francs = 6240 dollars / month. Assuming an American work week of 40 hours and 176 hours/month, that will come out to $35/hour. For an overtaxing, Westernized European country, that rate seems to be a very fair wage to me.
I think this will become more common in the future. Big corporations who *never* give a cent (I’m looking at you Cisco and others) but use OpenSSH (for example) could sponsor the development of OSS, not necessarily Linux. I personally would love it if all the commercial OpenSSH users donated some cash to the OpenBSD guys. The battle against the proprietary Microseft monopoly continues.
You have rather limited hw resources on this embedded machine.
Maybe there were some major changes in 5.x, that increases scalability only on faster processors.
What about benchmarking inside on LAN, where you have greater control over your network? You should also use memory-to-memory transfer of data, without involving the filesystem itself.
Benchmarking is a rather complex tasks, one can easily misunderstand the results.
> You have rather limited hw resources on this embedded
> machine.
> Maybe there were some major changes in 5.x, that
> increases scalability only on faster processors.
>
anything much faster than that is fast enough that you can’t notice the difference, i.e. isn’t CPU bound at a full 100 Mb (on 4.x, 5.x, or 6.x). During gig testing on faster hardware (mid-high end P3’s and low end P4’s), the same is apparent.
> What about benchmarking inside on LAN, where you have
> greater control over your network?
as I said, that’s a very limited subset of all the testing I did. I did do LAN testing, testing between two systems with nothing but a crossover between them, etc.
As a result of all of the testing I did, some performance results from others, and an email exchange with Robert Watson that basically said they knew it was an issue and were working to help improve things, m0n0wall switched back to 4.x for the time being.
It’s a substantial amount slower with a single TCP stream, and even slower as you start pushing through more and more TCP sessions.
With some network performance improvements in 6.0, rwatson said he’d seen reports of 20% improvements in throughput on SMP systems, though it “isn’t enough to recover full 4.x performance”.
Robert said their goal is to recover “almost all” of 4.x’s network performance on single proc, and substantially outperform 4.x on SMP, and I’m sure this work is a critical portion of that.
Of course if you want to use the machine for these tasks, its clear that 4.x will perform well.
As far as I’m concerned the amount Andre is asking for such work extremely reasonable. What I’d like to know from other BSD developers is what his relative reputation for code quality and project completion is. Not being a programmer, I can’t do that myself by just looking at his CVS commits. Assuming that he is capable, competent, and dependable I think his offer is extremely generous!
How sad someone (expecially if his/her job is worthy) is forced to drop his dignity and go out asking for money from door to door.
And how sad people can quietly discuss if he’s worth 20$ or 35 or if he needs a dishwasher or not.
Not to mention listing people who sent you 20$… and yet there are people arguing he cannot keep up with his promises for a 50% improvement. Yeah, maybe he could get his 20 bucks anyway if he will sing a nice song or put up a red nose.
This is not how I imagine an human being running his/her life and it reminds me when poor hungry people had to ask for permission to hunt a deer in king’s countries.
Now let’s all turn our projects into OSS software and let’s all have a blog where we post dimes we received form Mr. Jobs or Mr. Gates or Queen Elisabeth.
I hope this post will wake up people before they dig their own grave.
Sorry, but this is by far saddest thing I read in a long time. I didn’t mean to troll but I wouldn’t release anything as OSS anymore. I don’t know if he’s to blame but I would rather starve rather than writing something like this. I didn’t think I would ever read something like this in a World where Marx was born. Sad.
What do you mean drop his dignity? He actually begging. He doesn’t have to take time of his job. He is simply offering his time to the community for a price. He too has got to live you know. Some one said that at 40 hours/week the price he is taking is reasonable. If you read the link he is in fact proposing 60 hours/week. Honestly I don’t understand your post at all.
Oops, I’ve just been trolled … nice.
Excuse me. It should have read “He isn’t actually begging”. What a difference.
TBPrince writes..
“How sad someone (expecially if his/her job is worthy) is forced to drop his dignity and go out asking for money from door to door. ”
??..
Are you insane?
The guy is just saying: if you pay me, I’m gonna do this and that. Sounds like a work proposal to me.
All the same, the “begging” thing (“asking for money from door to door”?..) and the “dignity dropping” are pretty much just in your head.
Time for M0noWall to move to DragonFLY…
What he seeks in sponsorship is a drop in the bucket for publicly traded companies.
Perhaps Yahoo! should consider sponsoring him. After all, don’t they use FreeBSD extensively and stand to benefit from his work?
While I believe the 40% performance improvement being quoted would be for SMP systems (due to the fact that FreeBSD networking doesn’t currently scale well), there is still opportunity for large increases – even faster than 4.x levels as shown by the below post.
http://marc.theaimsgroup.com/?l=freebsd-cvs-all&m=106741733526914&w…
Linux 2.6 has gone on to allow even more parallelism in the network stack while also retaining or even improving its single threaded performance.
How sad someone (expecially if his/her job is worthy) is forced to drop his dignity and go out asking for money from door to door.
I think it’s a good idea, personally. I’d much rather have established OSS developers with a track record of excellent work offering their services in this manner than not contributing because they can’t justify the time.