Upload/Download ratio for BC cache?

[Const2k]

Hi there.

Couple of days ago I gave BitComet 128MB of RAM for cache. It consumed all of them eagerly :) And some time (10-12 hours) later I came to check logs and found out that there had been power failure, system got rebooted, and BitComet resumed it's tasks. So, I'd like to estimate loss of downloaded data.

(I really like this configurable cache - it reduces I/O to ~1 in 10 sec for me... Why boasting how low your X client is on system resources? BitComet will take as much (or as little) as you give to it...)

...and I don't know where to start from. So here is some info on my connection:

24/7 online, 320kbit Cable connection (35-40 kB/s Down, 100-200 (up to 500) kB/s Up. Free memory always stays more than "Min.=50MB"

Any ideas?

Thanks beforehand.

[bitdave]

320kbit Cable connection (35-40 kB/s Down, 100-200 (up to 500) kB/s Up

I think you have it backwards here. I dont know of any cable connections that have a higher upload speed than download.

Have you set your Global Max Upload Rate properly.

Now I dont really know what your question is, but let me just say that the 128 MB of cache you had was not all stuff that needed to be written to the HDD. Some of that 128 MB were pieces that had already been written to the drive and were held in the cache for uploading purposes. There's no way to know for sure but you probably lost about 64 MB from that power failure. This should be no big deal though, what i would do is stop those torrents and run a manual hash check on them to make sure there is no corrupted pieces. If there is then the hash check will drop those pieces and they will need to be redownload. 64 MB shouldnt take too long to get back on your cable connection.

If that didnt answer your question please clarify what exactly you would like to know.

[Const2k]

Thanks for your reply, bitdave. Some further details....

I dont know of any cable connections that have a higher upload speed than download.

Have you set your Global Max Upload Rate properly.

Well, name it as you wish, it's UTP Cat.5E twisted pair CABLE with RJ-45 plugged into my network adapter ;)

Few times it gave me max.steady Internet upload of 500 kB/s. Download is limited by my ISP - 320 kbit/s = 40 kB/s. I think it's worth $20/mo I pay for it :)

...not that it matters...

I don't use upper limits as most of the times ACKs don't influence my downloads much.

My question is a bit technical: how is BitComet's overall cache divided between upload cache and download cache? What does it depend on (if it depends)?

"Statistics" gives me 0.1/s-0.2/s I/O operations' frequency and >95% "Hit Ratio". That's quite nice for me as HDD is accessed much less than it could be (thus prolonging its lifetime). I thought cache sizes are proportional to traffic speeds. And that's why I gave you all some numbers to start with.

I hope someone of BC stuff would be kind enough to throw some light here ;)

I want to know that because

1) I want to be able to estimate possible data loss in future and

2) I want to choose optimal size of cache for my machine (concerning results from 1).

Also, uTorrent claims it's low on resources; fine - let them use it if they want; I'd prefer more stability for my fileserver at cost of RAM usage. On another hand, I don't want to lose too much data "in case". I think it's acceptable for me to lose less than I can get for 15-20 min of downloading (~35MB in my case).

So... GIVE ME NUMBERS, SOMEONE! :)

what i would do is stop those torrents and run a manual hash check on them to make sure there is no corrupted pieces.

Hmm... I didn't think about it since I have auto-hash check on task completion enabled (it needs to be on - I'm AFK most of the time).

Question is still there...

P.S> Maybe I should've repeated it in my first message, not just leaving it in topic's name...

[bitdave]

Few times it gave me max.steady Internet upload of 500 kB/s. Download is limited by my ISP - 320 kbit/s = 40 kB/s.

Wow that is a really wierd internet connection, is this some sort of business connection.

I've never heard of anyone being able to upload at almost 4 Mb/s and then only get 320 Kb/s for downloads.

When i say upload I mean your ability to send out data and download means your ability to receive data.

Your speeds seem quite the opposite of most consumer cable connections, although maybe Moscow is different, but i doubt it.

If your upload really is such then likely all the stuff in the cache were pieces held for uploads and that means you didnt lose hardly any data at all. But i still think you have your download/upload speeds backwards.

I don't use upper limits as most of the times ACKs don't influence my downloads much

What do you mean? The only way you can download anything is if you have upload room for the overhead involved in downloading. If you dont cap your Global Max Upload Rate then Bitcomet will eagerly max out your upload transfers and then there will be no room for this overhead necessary for downloading.

Sorry I cant give you any concrete numbers, only Bitcomet's developer RnySmile would know how the cache is implemented.

PS: uTorrent also has a data cache and you can adjust how much data you want held for uploads and for downloads individually, its in the advanced settings of the stable version and in the newer betas it has been reworked and has its own option now.

[Const2k]

Wow that is a really wierd internet connection, is this some sort of business connection.

I've never heard of anyone being able to upload at almost 4 Mb/s and then only get 320 Kb/s for downloads.

That puzzled me too, but then I thought someone of leechers was near myself - because most of the times my Internet UL stays at ~150kB/s. Yet I DL ~70GB/mo avg and UL over 150GB/mo (stats for Internet only). An ordinary "unlim" connection here, and not the best I can get... Neither I nor my ISP seem to be anxious about uploading (maybe that's because I'm unaccessible from outside as I have dynamic IP and I'm behind NAT) - ISP limits what I need of it/him - my downloading from Internet, and I pay it/him just for that (I get 100Mbit LAN of my district free of charge, too).

[/me is away to spend some time on AT&T site]

Well, now I see why my numbers sound confusing... You have twice more speed for the same amount of $s than me. (here's online translation of my ISP's tariffs, if you're interested. 1"c.u"~$1.1).

What do you mean? The only way you can download anything is if you have upload room for the overhead involved in downloading.

Well, you got me, those two parts of sentence are unrelated :P

1) "ACKs don't influence": ACKs ON UPLOADS don't influence my download much. I started new task of seeding (someone made torrent out of files I have already had), and after a week or so got this:

As you can easily calculate, my download is about .2% of upload. Even if we suppose all this data is ACKs (no traffic between my comp & tracker, no DHT, PEX etc.)... I don't think download ACKs are different either :) so total ACKs would be .4% of traffic. Nothing worth speaking of.

2) "don't use upper limits": You say:

If you dont cap your Global Max Upload Rate then Bitcomet will eagerly max out your upload transfers and then there will be no room for this overhead necessary for downloading.

I thought so too, and had experimented on some well seeded torrents. I've got (roughly) that my DL is about 1/2 of my UL _for _this_task_. So when I upload 40kB/s (limiter on) I get 20kB/s; UL=80 => DL=40; when I UL more, I still DL these 40 kB/s... So why limit others' download, when I gain nothing from that? Large cache compensates increased HDD I/O (I would say "/O" :) )... Looks like I have separate "pools of traffic" for UL & DL, not like DSL users ;)

Well, I think the discussion is over, as BC stuff doesn't seem to respond... Nevertheless, knowledge that someone has the same idea on UL/DL cache ratio as me makes me calm :)

Viva la cache! Long live the HDD! :D

BTW, thanks for uT reference, I'll go check this out there. It's fun to have some competitors to BC as it makes BC improve faster :)