Martin Geisler <[EMAIL PROTECTED]> writes: > I am currently installing VIFF on the viff.dk machine hosted at > DreamHost (in Los Angeles in believe), USA. But it takes quite a while > to compile libgmp, gmpy, etc. since the machine is very busy (the load > is ~25!). I'll post some results when I get them.
The DreamHost install worked and I have now tested VIFF on three
machines placed on three different networks! My machine is hbox[1] in
Sabro, Denmark, the DAIMI machine is camel11 in Aarhus, Denmark, and the
DreamHost machine is viff in Los Angeles, USA.
Using traceroute, I made the following table with the hops and round
trip times between the three hosts:
+-----------+-----------+------------+---------+
| from \ to | hbox | viff | camel11 |
+-----------+-----------+------------+---------+
| hbox | --- | 10 / 197ms | ? / ?ms |
+-----------+-----------+------------+---------+
| viff | 7 / 175ms | --- | ? / ?ms |
+-----------+-----------+------------+---------+
| camel11 | 9 / 30ms | 10 / 198ms | --- |
+-----------+-----------+------------+---------+
The DAIMI firewall seems to eat the packets sent out by traceroute, so I
could not get any numbers for incoming connections to camel11. But
outgoing connections worked fine.
The table tells us that it takes a little less than 100ms to cross the
Atlantic and that the connection between hbox and camel11 is
significantly faster.
I tried the multiply_benchmark.py script with 1000 multiplications, and
this is the result:
Time per multiplication: 4.185 ms
Running the comparison benchmark with a large number of comparisons
proved difficult because DreamHost kills heavy processes running on
their systems. So the benchmark was killed repeatedly when trying 100,
50, or even 25 comparisons. But I did manage to do 10 comparisons:
Time per comparison: 1245 ms
The load on viff was about 3.5 when the test was run. Getting access to
a machine where we can do more comparisons would be interesting. I did
run the small test several times, though, and the numbers were
consistently around 1200 ms.
All tests were run with the standard 65 bit modulus and using revision
c1622b3dc6c4 of VIFF (current tip). Python 2.4 was used on all three
machines. I now know that VIFF is incompatible with Python 2.3 because
of the use of decorators. I think I like them too much to change the
code to be Python 2.3 compatible, but please let me know if you have an
issue with this.
[1]: way to much LaTeX for me... :-) But I do look forward to the day
when the disks on hbox become full.
--
Martin Geisler
pgpDHlOxZwjBN.pgp
Description: PGP signature
_______________________________________________ viff-devel mailing list (http://viff.dk/) [email protected] http://lists.viff.dk/listinfo.cgi/viff-devel-viff.dk
