Archived
This forum has been archived. Please start a new discussion on GitHub.
Performance measures for IceStorm
Has anyone done (and can share!) performance measurements on IceStorm?
Things like:
(a) throughput
(b) latency
(c) effects of loading
would be very useful!
Things like:
(a) throughput
(b) latency
(c) effects of loading
would be very useful!
0
Comments
-
Re: Performance measures for IceStormOriginally posted by SteveWampler
Has anyone done (and can share!) performance measurements on IceStorm?
Things like:
(a) throughput
(b) latency
(c) effects of loading
would be very useful!
i modifed ice def in IceStorm sample as following:
sequence<byte> xbuffer;
struct Message
{
string hostid;
long timestamp;
xbuffer buf;
};
interface Clock
{
void tick(Message msg);
};
before sending, Message.buf is set with 512 bytes.
tick()is called every 30ms, as one way.
IceStorm runs on p4 3.0G(HT)+1G Ram + 1000M NIC, client host run 1~5 publisher & subscriber.
total publisher + subscriber/latency(publisher->server->subscriber)/IceStorm cpu load
<5+5/0ms/<1%
5~20/1~3ms/1%~5%
20~30/100~8000ms/<17%
i think ipstack is full loaded when we have 30 subscribers & publishers.
when i get 1000M nic, i will run test on redhat ES to see the difference.0