Opened 8 years ago

Closed 7 years ago

#5038 closed task (fixed)

Measure obfsproxy/obfs2 performance

Reported by: arma Owned by: asn
Priority: Medium Milestone:
Component: Archived/Obfsproxy Version:
Severity: Keywords:
Cc: arma, ln5, skep, twilde Actual Points:
Parent ID: Points:
Reviewer: Sponsor:

Description

Once we get the client-side of #4927 (tbb + obfs) building, I wonder if we should seed it with a fast stable obfs2 bridge. That way you can use your own obfs2 bridge, but you can also try out obfs2 without needing to find your own.

Clearly, if the bundle becomes way popular, that bridge will get hammered in terms of bandwidth.

But will obfsproxy hold up until that point, on the bridge side? Or will it fall over earlier? Any guesses, or better, some way to find out for reals?

Child Tickets

TicketStatusOwnerSummaryComponent
#5059closedasnOrganize an obfsproxy/PTTB test runCircumvention/Pluggable transport

Change History (7)

comment:1 in reply to:  description Changed 8 years ago by asn

Summary: how many clients can a given obfsproxy bridge handle?Measure obfsproxy/obfs2 perfromance

Replying to arma:

Once we get the client-side of #4927 (tbb + obfs) building, I wonder if we should seed it with a fast stable obfs2 bridge. That way you can use your own obfs2 bridge, but you can also try out obfs2 without needing to find your own.

Clearly, if the bundle becomes way popular, that bridge will get hammered in terms of bandwidth.

But will obfsproxy hold up until that point, on the bridge side? Or will it fall over earlier? Any guesses, or better, some way to find out for reals?

I doubt I can accurately answer the first two questions. As I see it, obfsproxy (with obfs2, at least) should hold more load than tor, since its architecture and functionality is much more lightweight. Let's do #5059 and figure if this is true.

comment:2 Changed 8 years ago by arma

Summary: Measure obfsproxy/obfs2 perfromanceMeasure obfsproxy/obfs2 performance

comment:3 Changed 8 years ago by asn

Cc: arma ln5 skep twilde added
Status: newneeds_information

We did #5059. What did we learn about obfsproxy/obfs2 performance?

From what I've learned, obfsproxy is quite lightweight. Do we have any more specific data (like, the increase of RAM/CPU usage with the number of connections, etc.)? We might also want to use such data to evaluate the performance side of #5550.

(CCing people who had bridges on the bundle.)

Thanks!

comment:4 Changed 8 years ago by skep

I can't provide specific data, but the overall perfomance seems quite good. I haven't notice any big (or even smaller one) increase of RAM/CPU usage. It only goes up a bit if the bridge sees lots of traffic, independently from the number of connections.

comment:5 Changed 8 years ago by arma

So far so good here too:

tord2    10780  0.2  0.0  21264  4492 pts/15   S+   Mar27  58:24 ./obfsproxy

I think we can close this ticket. Its main purpose was to uncover issues like #5074, which we did.

comment:6 in reply to:  5 Changed 8 years ago by asn

Replying to arma:

So far so good here too:

tord2    10780  0.2  0.0  21264  4492 pts/15   S+   Mar27  58:24 ./obfsproxy

I think we can close this ticket. Its main purpose was to uncover issues like #5074, which we did.

I would say that #5059 was to uncover issues like #5074.

I considered this ticket a way to get performance statistics for obfsproxy when it's used by lots of clients (which would be useful to compare with #5550, or with future obfsproxy perfromance stats).

In any case, if we don't have better statistics than what we got, we can close the ticket. If we ever need better statistics we will probably find a way to get them.

comment:7 Changed 7 years ago by asn

Resolution: fixed
Status: needs_informationclosed

I guess this ticket is done.

Note: See TracTickets for help on using tickets.