tomcat-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Christopher Schultz <ch...@christopherschultz.net>
Subject Re: Configuration question for 2500 simultaneous users.
Date Sat, 03 Aug 2013 02:24:06 GMT
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

Alec,

On 8/1/13 6:14 PM, Tomcat Random wrote:
> I'm expecting 5000 simultaneous users, with a physical load
> balancer to two physical app servers. So ~2500 per machine, each
> running an instance of tomcat not fronted by httpd or any proxy
> server. (i.e., using Tomcat to serve a few static assets along with
> the webapp).

Again, how many simultaneous /connections/ -- or requests -- do you
expect to handle? The number of simultaneous users is really only
relevant if you use sessions and then it comes down to memory (heap)
and failover (replication) if you want it.

What's much more important is the transaction rate. For instance, if
you expect 5000 users (really ~2500 per instance) and they each make a
request once per minute, than that's 2500 requests per minute or ~40
req/sec. If your average transaction takes longer than 1/40 sec
(250ms), then you are in trouble because you will not be able to keep
up with demand.

If your users make 2 transactions per minute, then you need to have
average transaction time down to 1/80 sec (125ms) or you will not be
able to keep up.

What happens if one host fails and the other one picks up *all* the
traffic? That means you need to do the average transaction in 1/160
sec (62ms) or you will not be able to keep up.

You might consider using a hot-spare or 3 instances.

> "are you just interested in speculative performance tuning?" That's
> correct - with just one user (myself) testing it everything is
> wonderful. I've just got that uneasy feeling the servers will
> explode when we flip over from our old host/codebase and get all
> the traffic.

JMeter is your friend. Put in the time to build workflow simulations.
It will pay you back manifold. You can even use your JMeter load tests
to do quick smoke-testing in production after a release just to make
sure you didn't miss anything.

- -chris
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.14 (Darwin)
Comment: GPGTools - http://gpgtools.org
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQIcBAEBCAAGBQJR/GnEAAoJEBzwKT+lPKRYx68QAJPE1VHVN4tX+cmssoaO3Qhr
4zjRtYYmpYna8L4EfAjPX8E3/owXilFj5bLBEpl6eYJg+i+RF+8+zlut45OXqrMO
1s5pvgwglq6kzZw70Wt8PavxP9ppcp2kwy82rNQbDLaskfkGcsV2o9QnJzcNgIIC
1Kr7jf1iqF7qJC8/F9F3cg3YBvzz2fozdaX1nWigdus9QsXq00ZRjByR1rd43ePH
gLEoJVzD95djWYP/JRiv65ZTqdDY6dMMsHwlfSfFPa9/w2uoZsjFaJmg8Eoj0K9o
4f2RHzIioizY9pLPWqRh4D3rOBOlTKMZyiRrYh4gdVpYWfS6D0Ae5zZx67r0VFje
MdVM3AUUyADwsHIE3+MBRu/OAeYnMDHIMHRlXON/YNXwYpRYMI9deS2y+kkYcxhY
ttl+B88UBRD0MR5twmCJ2OAAMe5WNtq0jlhuAGTnPJpDHSSd6fa/sm8Csoj1gdqr
ZSTh0JUy6RlECePeVHW+FCEBDPsWfM05pHY8p73KZrS3c4QWPXsH5GmnxGOcOuQh
Az1agFN4gqebFTG6l1K40CBSf7DWw38aYw+sFZWLJLM5/eSQtIaYgKhMwOu1N0Cp
oVgtwRVWQ0x/l3B4WNfgMHDq5xigD+ElGcvhPhRepzHPH4dZJxLYlpHDSV31o6k+
Myi3cewJzt+JVL4KN1yf
=u5fy
-----END PGP SIGNATURE-----

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
For additional commands, e-mail: users-help@tomcat.apache.org


Mime
View raw message