About :: TODO :: Blog :: RSS :: Old blog :: Projects :: GIT :: Gallery :: Notes

Thu, 30 Nov 2006

New kevent 'take26'release.


Short changelog:

• use struct timespec as timeout parameter.
• added high-resolution timer to handle absolute timeouts.
• added flags to waiting and initialization syscalls.
• kevent_commit() has new_uidx parameter.
• kevent_wait() has old_uidx parameter, which, if not equal to u->uidx, results in immediate wakeup (usefull for the case when entries are added asynchronously from kernel (not supported for now)).
• added interface to mark any event as ready.
• kevent POSIX timers support.
• return -ENOSYS if there is no registered event type.
• provided file descriptor must be checked for fifo type (spotted by Eric Dumazet).
• signal notifications.
• documentation update.
• lighttpd patch updated.

/devel/kevent :: Link / Comments (0)

Hoax or not? New kevent benchmark.


I've completed a bunch of kevent updates including major kevent interface changes and will release new version soon. I've also created some locking changes, which I call speculative locking (i.e. some fields (ret_flags mostly) can be updated without locks, but it should not harm the system), which can be wrong, but results are impressive.

Hardware setup.
Client: Intel Core Duo 3.4 Ghz (run with 3.7 Ghz each core), 2 GB of RAM, sky2 driven gigabit NIC (Marvell 88E8053).
Server: AMD Athlon 64 3500+ 2.2 Ghz, 1 GB of RAM, r8169 gigabit NIC.

Software.
Both client and server systems have this network timewait socket setup:

# echo 1 > /proc/sys/net/ipv4/tcp_tw_reuse
# echo 1 > /proc/sys/net/ipv4/tcp_tw_recycle
# ulimit -n100000

  ab -n80000 -c8000 http://192.168.0.48/

kevent about 7200-7400 requests per second
epoll about 4300-5300 requests per second

kevent about 4000 requests per second
epoll about 3800 requests per second

/devel/kevent :: Link / Comments (0)

Grange's song on OpenBSD hardware mini-hachathon h2k6.




Here is mp3 version (6.1 Mb, renamed to moscow.mp for those who has corporate filters for mp3 files) of the song from Mongol Shuudan band. This song and some others are available from band's homepage - mostly punk style.

/devel/other :: Link / Comments (0)

My J. Michael TR-300S trumpet has arrived.


I have not tried to extract sounds from it yet - I'm in office and I want to get home without major damages, which are unavoidable if I would try to 'play' right now - there are several people here, and they even try to work, so my first lesson is postponed to the evening.

I will start to rape and crack neighbours ears later today at home - likely no one will suffer too much, building is quite new, so there are only about 20 'alive' flats in the whole building (about 350 apartments).
I think my sounds will be in a harmony with sounds extracted from concrete breaking perforators and corner-grinding machines.

/life :: Link / Comments (0)

Wed, 29 Nov 2006

Perfect. Just perfect.


I'm talking about my climbing training, if you have not understood.
Actually there were nothing special from climbing point of view - the same traverses and bouldering, some progress here and there, bits of mixed traces and so on. But the whole training was extremely good - when interesting people smile to you and you smile to them, when you found new moving on the old trace, when body aching after the holds, when ... - all that small bits create excellent mood and I become to think that I just spent perfect time, and actually most of the time (I would even say overwhelming part of the time) I spent was really good.

/life :: Link / Comments (0)

OpenBSD hardware minihackathon is over.


Grange has really burnt them brains - in the brazilian pub he took away a guitar from singer and started to sing russian song "Moscow" with Sergey Esenin words.
I would say that Grange signs not that bad, and especially when he is drunk, and song itself is very good too, but the whole situation made it even better, so he was called 'the most talented man at h2k6'. It is quite possible. One can even find his records on the homepage (in russian).
So, it was very interesting meeting, as we can see.
Among good mood he also got bunch of various hardware, so I hope he will finally start hacking interesting stuff instead of slacking his ... fifth point.

/devel/other :: Link / Comments (0)

Multidimensional trie implementation.


Although long-running test was completed without errors (actually it is not interesting test at all, since 10 rules is a too small set), after some thoughts I've found a bug in the algorithm - I tried to avoid recursive searching in case of wildcard nodes found traversing trie, but it seems that it is unavoidable.
I also added trivial memory leak detector, which found another bug in algorithm - there is possibility to not free next-dimensional roots in some cases.
So I started to change algorithm itself, and although its theoretical ground is fully completed, practical implementation has a place where bug lives - I know exactly what is the bug and where is that buggy place in the code, what are the reasons, actually I know everything to fix it, but do not know how to make a good fix. Yet.

/devel/networking :: Link / Comments (0)

Tue, 28 Nov 2006

Multidimensional trie implementation has been completed.


It is a first and the most significant step in netchannels wildcard implementation. I tested it with different number of dimensions (1-5) and different number of added/searched/removed rules (upto several millions in each dimension).
Algorithm is simple and fast - my previous estimation was incorrect - currently 1 million 5-dimensional insertions take about 6 seconds, 1 million 1-dimension insertions take about 1.1 seconds on my test machine.
When I add search/remove operation (one-by-one, i.e. add one, search for it and remove it), the whole tuple takes about 15 and 3.2 seconds for 5-dimensional and 1-dimensional case accordingly.

I will (probably) create some graphs describing algorithm and how it will be used with netchannels tomorrow, but if algorithm itself is implemented correctly, it is the most significant part of wildcards support fot netchannels.

I have started following testing case for 5-dimensional case for the night: 10 rules are added, then 10 various values (actually inserted values, but with 0xffffffff mask) are searched for, and then all they are removed, let's see how this will end up tomorrow.

/devel/networking :: Link / Comments (0)

New musical toy.




P.S. Do not scream and laugh, but I can not play trumpet - will start to learn.

/life :: Link / Comments (0)

That is what I call good hacking.




Other photos from h2k6 hackathon.

/devel/other :: Link / Comments (0)

Mon, 27 Nov 2006

Climbing evening.


After quite empty day (I did not hack, did not work, did not do anything usefull) I moved to climbing zone and spent there excellent time - there were several old traverses completed, couple of new boulderings. The most interesting part was relief traverse - each day I add more and more holds there, although it is marked not as the most complex trace, I only saw one person who tried to work with it (except me), others somehow do not like that trace. unfortunately in Skala-city there are no relief vertiacl traces, which I'm quite sure would be my favourite ones.

/life :: Link / Comments (0)

New Year is coming.


I've started to think about presents.
First of all, present for myself (well, second one actually) - I think I want to start to make photos, so I decided to get Cannon EOS 400D. Not sure if I will really buy it, but at least I think about it.
The best present for myself would be flat repair complete, but we will see...

So, to touch the perfect, I've subscribed to couple of interesting livejournals of photo masters.
There are indeed several types of photos I like - first of all, it is night photos, then probably building/stuff photos (especially from higher floors or roofs) and panorams.
I have extremely small experience in that area, but will try to learn. Maybe.

/life :: Link / Comments (0)

OpenBSD h2k6 hardware mini-hackathon.


Day first (in russian).
Grange did not forget bottle of "Russian Size", but it was seized in Zurich.

As became known from authoritative sources, Grange tries to add support for IBM ServeRAID to OpenBSD.
It looks roughly like this:


Small photo archive of the meeting.

This algorithm describes what is hackathon and what they are doing there:


For comparison: Linux network developers conference NetConf is completely different meeting - hackers sing karaoke songs and seems to not drink at all, I never participated there though so can not say for sure. On picture core linux network hackers (David Miller, Rusty Russell and not-yet-a-netdev-hacker Soyoung Park) select karaoke song.

/devel/other :: Link / Comments (0)

Sat, 25 Nov 2006

Initial trie implementations for netchannels wildcard support.


I completed 1-dimensional trie implementation and started its wild testing - random prefix and address mask is created in infinite loop and then random address is being searched there - it has passed more than 123 millions insertion/searches and ate about 1.8Gb of ram - each trie node is 20 bytes in size, which would take 2.29 Gb of RAM if they lived as-is, but due to the nature of tries, which allows to reuse the same entries for several prefixes, memory usage is smaller. 100k insert+search operations took always less than 2.5 seconds on my test machine.

Next step, which I will start after weekend, is multi-dimensional trie implementation, which is actually netchannel wildcard support.

/devel/networking :: Link / Comments (0)

Grange has flown to OpenBSD hardware mini-hackathon.


This small party, also known as 'I-can-drink-more-than-you-can-hack' (there is no real evidence that they spend there time with profit for the world or at least for own health), will be spent in University of Coimbra in Portugal.
Grange will hack there probably Dallas one-wire bus, maybe something else. Actually I do not know, why he is going there except that to become drunk with people which live about tens of thousands kilometers away.
I hope he did not forget to get one bottle of 'Russian Size'.
So, let's wish them nice time there.

/devel/other :: Link / Comments (0)

Fri, 24 Nov 2006

The main development words.


Rusty Russell's "Hanging Out With Smart People".

/devel :: Link / Comments (0)

New books.


I bought yesterday evening two books about history of two interesting states:

• "History of Russia. From the ancient times to 21 century." A. N. Saharov (Russain Academy of Sciences).
• "All under heaven. A complete history of China." Rayne Kruger.

/life :: Link / Comments (0)

Kevent last words.


Ok, after I was requested to change interfaces again to the state they were in the previous release, I've thrown the last words - I'm waiting for about one week and listen for other developers to comment on the interfaces, if no one will comment, I will gather all the latest suggestions, implement them and make the new release - no interface changes after that day.
And I do not care if there will be some problems and/or some people do not have time to review all my releases or something else - I have time to implement them on per-week basis (I could create new release every day actually), so there is plenty of time to review for those who cares.
If you have something to say about kevent, your opinion on its inclusion/declining or on my behaviour and so on, it is the right time to say.

Thank you.
I'm going to implement netchannels wildcards - it is much more interesting than spending hours on completely void tasks.

/devel/kevent :: Link / Comments (0)

New kevent features scheduled for the next release.

• struct timespec and possibility to have absolute timeouts in syscalls instead of simple number of nanoseconds.
• signal mask parameter for syscalls to allow temporal blocking of the signals while syscall is executed (almost, but I'm still trying to show that it is stupid idea with kevents).
• kevent posix timers support.
• allow kevent to mark any events as ready from userspace.
• allow to break any waiting syscalls.
• various errno return values update.
• various interface changes, like addition of some new fields, and removal of otheres, which will not affect functionality behind that syscalls.
• remove some locks in the hottest path - it is indeed needed task. I plan to refactor searching/modifying code so that there were places in the struct ukevent which are never modified by another than caller users (like simultaneous changes in readiness state machine and searching part), so there would be possible to only have one lock in the readiness state machine around ready queue.

/devel/kevent :: Link / Comments (0)

Thu, 23 Nov 2006

Kevent posix timers support.


Early morning hack has been sent to mail lists for review.
I have not tested it (except compilation test), and frankly do not want, but if my flow of thoughts is correct, I will create simple application, which will use timer_create() and friends to test kevent posix timers support.
Implementation (and design of POSIX timers API) does not allow to have KEVENT and SIGNAL or THREAD delivery mechanism simultaneously, at least I concluded it after quick look into the code, which is probably is not correct.
Let's see if my patch is good or not.

/devel/kevent :: Link / Comments (0)

Wed, 22 Nov 2006

Climbing evening.


That was quite easy training - I ran several old traverses and couple of bouldering, also created one traverse, which previously contained green holds, but most of them were removed or replaced with time, so it was a new trace, that's all - nothing special this day.
The most pleasant part was sauna and shower which definitely were required by development related dirtiness.
Now I feel mysef just bloody excellent.

/life :: Link / Comments (0)

New features scheduled (implemented) for the next kevent release.


Most of them are implemented in the local tree, I'm just waiting for ack from Ulrich Drepper:

• new command, which allows to mark any events as ready. It is also possible to just wake up listener of the given kevent queue if number of events requested to be marked as ready is zero. It allows glibc to notify other event processing threads, which are parked in the waiting syscalls like kevent_wait() or kevent_get_events(), that they can process new events in cae of originally awakened thread was terminated.
• fixed unsufficient parameters check in kevent pipe notifications (spotted by Eric Dumazet).
• return -ENOSYS if there is no registered event type instead of -EINVAL.
• add 'flags' parameters to kevent_init() syscall (not implemented yet).

• just implement all horrible workarounds and ugliness I'm asked about, which I think will allow quite fast kevent integration into mainline.
• continuously perform seems to be endless negotiations with mutual insults in linux-kernel@. This will not force kevent inclusion, and likely will not end up with it at all, but there is probability (I think it is somewhere between zero and void), that I can convince people that I am right. Practice shows that it is hardly to convince people, which do not hear what you talk to them, no matter correct it is or not.
• say that 'I do not care and will not even discuss all requested utterly damaged ideas' or something like that and stop discussion. It can or can not help with kevent integration, but in any case I at least will have enough time for more interesting tasks than empty talks in linux-kernel@. For example netchannels wildcard support and fast NAT implementation, zero-copy support for netchannels, asynchronous IO, threading library, log-structured filesystem

/devel/kevent :: Link / Comments (0)

First part (about one fourth) of the hinged ceiling is completed.


Something like that.
I screwed it yesterday evening, and can say that although making hinged ceiling alone is not that simple task, it is quite doable.
Next task is to perform second layer, it will be lower and take more square, but it is task for another Sunday.



Will create better photos when it is ready.

/devel/flat :: Link / Comments (0)

Tue, 21 Nov 2006

New kevent 'take25' release.


Short changelog:

• new (old (new)) ring buffer implementation with kernel and user indexes.
• added initialization syscall instead of opening /dev/kevent.
• kevent_commit() syscall to commit ring buffer entries.
• changed KEVENT_REQ_WAKEUP_ONE flag to KEVENT_REQ_WAKEUP_ALL, kevent wakes only first thread always if that flag is not set.
• KEVENT_REQ_ALWAYS_QUEUE flag. If set, kevent will be queued into ready queue instead of copying back to userspace when kevent is ready immediately when it is added.
• lighttpd patch (Hail! Although nothing really outstanding compared to epoll).

And special blog edition changelog entry (interesting, do people read my patches to detect this lyrics? Let's see...):
• small three-line verse (I think I created not bad step for non-english speaking person) about one-line optimisation. Casted by Shakespeare.

/devel/kevent :: Link / Comments (0)

Hinged ceiling installation.


That is how I spent Sunday.
Weight of that part is about 20 kg, it was challenging to lift and fasten the detail and it is still in a hung state, since I do not have long enough screws. But I will complete it soon, maybe even today.

/devel/flat :: Link / Comments (0)

Mon, 20 Nov 2006

New kevent release.


It will be 'take25', but I have not released it yet, since it is in testing mode right now.
This release will include:

• new (old) ring buffer implementation with head and tail in userspace buffer with two syscalls to wait and to commit events (the latter becomes just an index update).
• additional feature of the ring buffer - overflow counter to prevent situation when two threads are going to free the same events, but one of them was scheduled away for too long, so ring indexes were wrapped, so when that thread will be awakened, it will free not those events, which it suppose to free.
• wake-all-threads flag - when it is set all interested threads will be awakened, otherwise only first one.
• initalization syscall and removed /dev dependency.
• small three-line verse (I think I created not bad step for non-english speaking person) about on-line optimisation. Casted by Shakespeare.

  ab -n50000 -c5000 http://192.168.4.78/

lighttpd_kevent:  4902.17 req/sec
lighttpd_epoll :  4843.71 req/sec

lighttpd_kevent:   4307.18 req/sec
lighttpd_epoll :   4027.34 req/sec

/devel/kevent :: Link / Comments (0)

New district and shops.


I live in a new district which is only became to be inhabited.
There is a bus station near the buildings and generally there are a lot of people there and obviously it is a good place for small shop or stall which would have cigaretts, beer and other such stuff.

And what would you think is the first shop opened there?
It is flavour shop!
I love my country.

/life :: Link / Comments (0)

Sat, 18 Nov 2006

New generic wildcard search design.


Well, it is not new, but it is different from what was described before. And although previous variant has higher lookup speed, it requires too much memory, so it is not acceptible.

New one is quite simple - it will be implemented on top of trie structures. Main feature of tries is that they are quite compact (in case of PATRICIA tries) and allow W^d lookup speed in the worst case, where W is width of the interval in bits (i.e. 32 bits in current netchannels implementation, which does not support IPv6 due to my laziness and absence of requirements feedback from the community), and d is number of dimensions (5 currently), which results in very bad number of 33.5 millions operations in the worst case of (2^W)*(2^d) == 2^(W+d) (137.5 milliards) rules.
Actually with netchannels it will be much smaller since only two dimensions use full 32 bit space (source and destination addresses), while others use either 16 bits (source and destination ports) or 8 bits (protocol number), so it will be something about 2 millions operations.

Basic idea behind tries is following: trie is a simple unbalanced tree each node of which represent only one bit (i'th level of the tree represent bits in i-1 position in the searched value), so it's maximum depth is 32 (in PATRICIA trie several bits can be combined into the same node if there are no other rules, which have different bits in that positions). In that case rule 192.168.0.0 - 192.168.0.255 will be presented as 0b11000000101010000000000000000000(192.168.0.0) - 0b11000000101010000000000011111111(192.168.0.255), so actually trie will be stopped at bit 24 (counted from the most significant one) and the rest will present a wildcard (0-255 in the least significant byte in host order) or it will contain additional nodes if there are additional rules which has bits in that range, for example 192.168.0.0-192.168.0.20.
Each direction is a base for own tree.
Each node contains a pointer to the subtree of the next-dimension tree, where its rule starts. So when each node is examined (i.e. each bit is checked) in the first dimension and bit is found to to belong to some rule, next dimension is checked from the bit position specified in the node and this is being done recursively for each dimension. If in some dimension requested bit is not found in the trie (where it should be by the previous dimension rule), it means that there are no rules for requested values.

In the previous memory-hungry design, if rules do not have ranges but only single tuples (i.e. single source/destination IP address and single ports and protocol), or if all rules present the same range (for example all port rules have the same 1-65535 range), it can not lead to the described problem with memory usage. The more range overlaps rulset has, the more memory such algorithm will use - in theory, when each new rule requires interval split, i.e. when each new rule overlaps with all previous and has additional range (for example when each new rule is slightly bigger than previous one), it is O(d*W*(N^d)) where N is number of rules and d is number of directions, W is width in bits of each dimension (actually it can be hidden by O(), since it is a constant for different set of rules).

/devel/networking :: Link / Comments (0)

I wish my day would have 36 hours.


So I could quickly complete flat development, moved my laptop to HP repair (its disk fails after couple of hours of work) and started to work more than currently.

Likely if you do not see entry in the blog, that means that I'm too busy at paid work (or even unavailable at all) to work on my own projects.

/devel/other :: Link / Comments (0)

True asynchronous IO.


All existing AIO - both mainline and kevent based lack major feature - they are not fully asyncronous, i.e. they require synchronous set of steps, some of which can be asynchronous. For example aio_sendfile() requires open of the file descriptor and only then aio_sendfile() call. The same applies to mainline AIO and read/write calls.

My idea is to create real asyncronous IO - i.e. some entity which will describe set of tasks which should be performed asynchronously (from user point of view, although read and write obviously must be done after open and before close), for example syscall which gets as parameter destination socket and local filename (with optional offset and length fields), which will asynchronously from user point of view open a file and transfer requested part to the destination socket and then return opened file descriptor (or it can be closed if requested). Similar mechanism can be done for read/write calls.

This approach as long as asynchronous IO at all requires access to user memory from kernels thread or even interrupt handler (that is where kevent based AIO completes its requests) - it can be done in the way similar to how existing kevent ring buffer implementation and also can use dedicated kernel thread or workqueue to copy data into process memory.

It is very interesting task and should greatly speed up workloads of busy web/ftp and other servers, which can work with a huge number of files and huge number of clients.
I've put it into TODO list.

Someone, please stop the time for several days, so I could create some really good things for the universe.

/devel/other :: Link / Comments (0)

Fri, 17 Nov 2006

Climbing evening.


It was quite easy training, since there were crowds of people, which usually means, that it is quite problematic to climb horizontal traverses. But nevertheless I completed several old traces and started parts of the relief traverse (longer and longer each training, which is a good sign). At the end I also completed couple of start of complex traces on negative slope and found an interesting bouldering (read: lurked in climbing school training which happend near the place where I had a rest), but failed to complete it since was tired already.
Sauna, shower at climbing zone and a little beer with Gabriel Garcia Marquez book at home completed the day - yet another good day has gone...

/life :: Link / Comments (0)

Netchannels wildcard design and implementation.


It was first described here, but I did not present memory usage for that scenario.
After some calculations I found it is about O(d*(N^d)/4 + N) where N is number of netchannels in the whole tree and d is number of dimensions we search in. With 5 dimensions (source/desctination address/port and protocol number) it becomes completely unfeasible.

And after my implementation completed (although it requires some cleaning now) I've found that memory limit on practice.

Presented design allows extremely fast searching, but it is too memory hungry, so I will modify it to not replicate sub-trees of intervals in each interval, but instead to have different kind of trees which will reduce memory consumption down to O(d*N) instead of above O(d*(N^d)) and increased search time.
So, stay tuned...

/devel/networking :: Link / Comments (0)

Wed, 15 Nov 2006

Theatre evening.


I've seen Eugene Grishkovec's 'Po Po' performance created under expression of Edgar Po works.
Performance is a compilation of many novels written by famous author, all of them are easily detectible.
Grishkovec plays with Alexander Cekalo, they call it 'fun story about the death' and it is really so. I would even say that Grishkovec plays second role in that single-actor monologue performance.
There are some opinions that Grishkovec has star fever, but I would not say so - neither during his presentation before the performance, nor during the whole act he showed any signs of it, I managed to get tickets in the first raw, so I saw actors right in a couple of meters. Even after performance, when we met Grishkovec on the street with his girlfriend and asked couple of questions and presented a flower he behaved very good.
I like his performances, they are really different from the mos others, but unfortunately he plays quite rarely in Moscow. I hope I will see his 'Drednouti' ('Dreadnoughts') performance this season.

Later we got couple of beers in 'Vobla' beer restaurant and moved home. I think the whole day indeed succeeded, although it looks I caught a cold in addon to my current sick...

/life :: Link / Comments (0)

Packets in alternative worlds of netchannels.


Do you have an excellent spatial imagination?
Can you consider yourself and easily operate say in 3-dimensional world?
Likely you can, since we live in such a world. Probably we can think about 4-dimensional world too.

But can you describe to someone 5-dimensional world?
Or can you put yourself into one of the N worlds, where N is potentially millions, each of which has number dimensions smaller than previous world by one and upto N subworlds in each dimension? And while you - poor walker - are searching for the shortest way to the destination point in that brothel, someone looks into your universe through the glas and loudly lough counting your times clock.

I've just described couple of moments in a single network packet live, when it enters netchannels subsystem with this wildcards design.
It looks like it will have complex live, but we will see, since I've implemented generic multi-dimensional tree hierarchy and will complete searching algorithm and start testing tomorrow.

And now I want to finish my fantastic narration, since I'm going into a theatre.

/devel/networking :: Link / Comments (0)

Reading S. Minaev "Duhless" (aka "Spirit less").


It is extremely 'trendy' book, here one can find a chapter called 'Office'.

Crap. Complete crap - some kind of cool artistic slightly squeamish flood of mind(less).
One can probably think that it is trendy to read it, that it is cool, it is another view on our usual life and so on...
Maybe it is right, but I think it is a crap created to show 'how different author is from all those stupids'.

There are quite a few fun moments though - probably life of the top management really looks like it is described.
"What about putting the sea or beach or wild nature against a background of a woman".

But why so declamative? Why so depreciatingly? Crap.

/life :: Link / Comments (0)

Tue, 14 Nov 2006

Delayed allocation and fragmentation.


Quotation from Andreas Dilger:

Essentially, delaying the disk allocation until a file is large/complete
(delalloc) and then using a buddy allocator in memory to get contiguous
chunks of disk is better than hard 64K+ cluster because it avoids
internal fragmentation and allows much more optimal/efficient placement
than just a factor of 16 reduction in the block count.

/devel/fs :: Link / Comments (0)

Why trees are better than binary search.


For netchannels wildcard support I need to create set of non-crossing intervals which will be examined for each packet. It is possible to put them one-by-one and thus form an array of intervals which can be traversed using binary search, but there is a major problem when we will add new interval, which should be placed somewhere inside the existing array - that will force array reallocation and copy. This problem also applies to hash tables and actually all other table-based approaches, but does not exist in trees, although tree requres a bit more memory.
both have the same search speed about O(log2(N)), where N is number of intervals.

Rough code snippet (no return value checks and reference counters update), which splits intervals into non-crossing ranges:

	if (in->min < new->min) {
		netchannel_interval_split(in, in->min, new->min-1, GFP_KERNEL);
		if (in->max > new->max) {
			netchannel_interval_split(NULL, new->min, new->max, GFP_KERNEL);
			netchannel_interval_split(new, new->max+1, in->max, GFP_KERNEL);
		} else if (in->max < new->max) {
			netchannel_interval_split(NULL, new->min, in->max, GFP_KERNEL);
			netchannel_interval_split(new, in->max+1, new->max, GFP_KERNEL);
		} else {
			netchannel_interval_split(new, new->min, in->max, GFP_KERNEL);
		}
	} else if (in->min > new->min) {
		netchannel_interval_split(in, new->min, in->min-1, GFP_KERNEL);
		if (in->max > new->max) {
			netchannel_interval_split(NULL, in->min, new->max, GFP_KERNEL);
			netchannel_interval_split(new, new->max+1, in->max, GFP_KERNEL);
		} else if (in->max < new->max) {
			netchannel_interval_split(NULL, in->min, in->max, GFP_KERNEL);
			netchannel_interval_split(new, in->max+1, new->max, GFP_KERNEL);
		} else {
			netchannel_interval_split(new, in->min, new->max, GFP_KERNEL);
		}
	} else {
		if (in->max > new->max) {
			netchannel_interval_split(in, in->min, new->max, GFP_KERNEL);
			netchannel_interval_split(new, new->max+1, in->max, GFP_KERNEL);
		} else if (in->max < new->max) {
			netchannel_interval_split(in, in->min, in->max, GFP_KERNEL);
			netchannel_interval_split(new, in->max+1, new->max, GFP_KERNEL);
		} else {
			netchannel_interval_split(in, in->min, in->max, GFP_KERNEL);
		}
	}

/devel/networking :: Link / Comments (0)

Filesystem stress testing and emulation tool.


I wonder why Linux does not have FS stress testing and emulation tool like netem for network, which allows to perform drop, corruption, reordering, duplication and other interesting options extremely usefull for protocol testing.

Consider the same tool which works under FS layer, for example as ATA/SCSI controller, which can perform various message mangling, delaying, reordering and other features emulating incorrect behaviour of the hardware. It can even send data down to the real hard drive, thus emulating even various software conditions like scheduling starvation, irq storms and so on.

I think such tool is a must for fs development and should be implemented at least in parallel or even before various userlevel stress tools are started to stress the filesystem.

This deserves entry in TODO list.

/devel/fs :: Link / Comments (0)

Do you know what Google Earth lacks at the first place?


No, not 3d buildings - it lacks ion cannon.

/other :: Link / Comments (0)

Mon, 13 Nov 2006

Climbing day.


It was day of bouldering training - I tried different parts of interesting relief traverse and several old traverses over usual holds.
There are quite interesting moments when one needs to place his/her leg above the head to complete the hold and be able to get next one.
I think I like relif pathes more than any other traces, but unfortunately there are no vertical traces over reiefs in Skala-city.
Training was completed with campus-board exercises, sauna, shower and excellent mood.
I like it.

/life :: Link / Comments (0)

Ceiling design.


I think I will end up with something like this (handmade paintings with very bad quality).
If you feel yourself having enough spatial imagination, try this picture of the room I have in mind.



If you know some site where I can get an inspiration, related to my interests, do not hesitate and drop me a link...

/devel/flat :: Link / Comments (0)

Sun, 12 Nov 2006

Meanwhile on flat development front.


I completed to work with plaster - all walls and ceiling are ready, although they require polishing, which is quite fast procedure, but too dusty, so it will be done slowly.
I've found a way to create very nice corners in doorways using special fabric called 'serpyanka'.

Next step is to start hanged ceiling implementation in the room - I still have not created design of that monster, but I already completed electricity wiring for additional lights and neon cord.

I also decided to not put neither bath nor shower cabin into my bathroom, instead I will develop special podium for water sink and create some kind of a wall - it will be hand-made shower cabin 'on the floor', currently without glass dorrs, but I will add them eventually.

/devel/flat :: Link / Comments (0)

Sat, 11 Nov 2006

Added kevent signal notifications.


Signals which were requested to be delivered through kevent subsystem must be registered through usual signal() and others syscalls, this option allows alternative delivery. With KEVENT_SIGNAL_NOMASK flag being set in kevent for set of signals, they will not be delivered in a usual way. Kevents for appropriate signals are not copied when process forks, new process must add new kevents after fork(). Mask of signals is copied as before.

Here is output of the test program signal.c available in archive:

sig_handler, signo: 10.
Dequeued signal: 10, nodel: 0.
Dequeued signal: 12, nodel: 1.

  Erlang                   : 0.443 seconds
  C++/POSIX NPTL threads 1 : 4.374 seconds
  C++/POSIX NPTL threads 2 : 0.190 seconds

  C++/POSIX NPTL threads 1                        : 5083343 usecs
  C/POSIX NPTL threads                            : 4821212 usecs
  C++/POSIX NPTL threads 1 (-fomit-frame-pointer) : 4738057 usecs
  C/POSIX NPTL threads (-fomit-frame-pointer)     : 4430010 usecs

epoll (edge-triggered):   248408 events/sec
kevent (edge-triggered):  269282 events/sec
Busy reading loop:        269519 events/sec

  epoll (no EPOLLET):                  57428 events/sec
  kevent (no ET):                      59794 events/sec
  epoll (with EPOLLET):                71000 events/sec
  kevent (with ET):                    78265 events/sec
  Maximum (busy loop reading events):  88482 events/sec

int kevent_ring_init(int ctl_fd, struct ring_buffer *ring, unsigned int num);

struct kevent_ring
{
	unsigned int		ring_kidx, ring_uidx;
	struct ukevent		event[0];
};