Re: Performance Checking

First let me apologize for my mis-interpretation of the term co-located.  I
have in the past (during discussions of disaster recovery ) heard the term
used in reference to multiple sites.  That is not the case of the original
posting which I have included below Bob's response.

However, as you can see, he mentions four servers.  There must be some sort
of load balancing going on.

My posting was to show that server performance was not always the case for
poor system ( servers, network, etc.) performance.  If a round robin DNS is
not used then some piece of equipment is translating the "virtual IP"
address in the DNS to multiple "physical" addresses.  If his clients are
behind proxies, then the device could be flooding one machine and leaving
the others to idle.

If his hosting environment is experiencing bottlenecks in their networks
(and I doubt he is the only one using them) then that too could cause system
performance problems.  Single traceroutes are only the beginning of
determining that.  There are software programs that will query agents
running on client machines that can give graphs of time vs. load, but I felt
that recommending that was a bit early in the game.

I do not even have to tell you guys that a server with low memory or
substandard disk I/O system will make a poor performing system or that a
multi NICed server can sometimes handle requests better than a single NIC.

But I had to start somewhere and the suggestions I gave are inexpensive
first looks at performance.

These are "real world" hosting problems and would make for good discussion,
both in what was used to determine problems as well as resolve the problems.


----- Original Message -----
From: "Bob Maine" <maine(at)2alpha.net>
To: "Bryan Bateman" <batemanb(at)home.com>
Cc: "Mailing List Account" <digitald(at)digitaldaze.com>; <moe(at)maxim-ic.com>;
<hwg-servers-digest(at)mail.hwg.org>
Sent: Saturday, March 18, 2000 12:26 PM
Subject: Re: Performance Checking


> Bryan,
>
> I still don't think you understand what colocation means. Colocation has
> nothing to do with load sharing it is just means your server is physically
> siting at someone elses location sharing space, network hardware, and
> connectivity to the internet in a cooperative manor.
> The ip addresses are a totally seperate issue and just depend on how you
> have your server setup. One physical address for one machine or per
> network interface card the rest of the addresses pointed to that machine
> by the isp and handled via host tags,
> multi homing etc. It is very possible that there is one server, one ip
> address with no load sharing going on at all. Even if there was a load
> sharing scheme in place with multiple machines involved a trace
> route would, I think, only show which path through the nameserver the data
> is going for that particular transaction, so one trace would show (within
> the isp's system, the name server, the router maybe, and the actual
> machine serving the data? I am not a network guy so I don't know about
> that end of things, but I do have colocated servers and feal pretty
> confident that it has nothing to do in itself with load sharing.
>
>
     Subject: Performance checking
          To: Recipient List Suppressed:;;(at)mail.hwg.org
        From: Moe Rubenzahl <moe(at)maxim-ic.com>
        Date: Fri, 10 Mar 2000 14:45:27 -0800
 In-Reply-To: <20000308024325.NUAJ354.mta1@mallard>

I need a good way to look at the performance of our web site. We have
a small home page (35K), yet some people (notably one of our VPs,
connecting from home, using AT&T @Home and a cable modem) report our
site does not load as quickly as it should. I can't say for sure
whether he is wrong or right and have no real performance monitors in
place. I do sometimes see a pause with one page or another.

My question is whether anyone can recommend some site performance
services or tools or other ideas.

A little background:

- Site is http://www.maxim-ic.com

- We are running four Unix servers, co-located at Exodus, and they
are very lightly loaded. We serve about 13,000 PDF data sheets
(typically 50K each) a day and around a million static pages a month.
Apache is on one machine; database and Cold Fusion on a second; the
other two are loafing.

- We run Apache for our static pages and some of the site -- maybe
20% of traffic -- comes from Cold Fusion, serving from Sybase SQL.
Perceived speed problems are not limited to Cold Fusion-served pages.

- We use server side include to serve our page headers and footers. I
know it's not the best practice but figured that with our processors
so lightly loaded, we were OK.














> Robert
> http://parlorsongs.com
> Musical Reminiscences of years gone by.
> An amazing collection of popular and classical
> sheet music from the 1800s to the 1940s.
>
> On Sat, 18 Mar 2000, Bryan Bateman wrote:
>
> > Yes I know that is the traditional definition.  If that was the case
would
> > nslookup not show two IPaddresses.  One for each location.  How else
would a
> > client machine be able to find either/or environment?
> >
> >
> > ----- Original Message -----
> > From: "Mailing List Account" <digitald(at)digitaldaze.com>
> > To: "Bryan Bateman" <batemanb(at)home.com>
> > Cc: <moe(at)maxim-ic.com>; <hwg-servers-digest(at)mail.hwg.org>
> > Sent: Friday, March 17, 2000 7:53 PM
> > Subject: Re: Performance Checking
> >
> >
> > > Bryan,
> > >
> > > I think you are confused as to what co-location means.  Co-location is
> > > where your machine is housed in someone elses server room on their
> > > network.
> > >
> > > It does not have anything to do with load balancing, round robin or
> > > otherwise.
> > >
> > > (since I didn't see the original message I cannot comment further,
however
> > > I just wanted to clarify co-location).
> > >
> > > On Fri, 17 Mar 2000, Bryan Bateman wrote:
> > >
> > > > I think you may want to examine the performance of the routes.
> > > >
> > > > Look at the traceroute below.  Seems like a lot of hops to me.  Also
> > note
> > > > that network response times go up by a factor of 4 on the exodus.net
> > > > network.  I recommend having someone check this during peak traffic
> > times.
> > > > You will probably see worse bottlenecking than you see here.
> > > >
> > > > You also mentioned co-location.  Is this being accomplised by
hardware
> > at
> > > > exodus.net's site or DNS round-robin.  Compare an nslookup of your
site
> > to
> > > > that of microsoft below.
> > > >
> > > > As you can see, you are resolving to only one IP while microsoft
> > > > round-robins to multiple IP addresses.
> > > >
> > > > Is there a network segment between exodus.net and your site?  That
is
> > the
> > > > only way I can see co-location occuring.  If so, you probably want
to
> > > > examine that network segment as well.
> > > >
> > > > All Internet traffic "appears" to be going to one location.
> > > >
> > > > If they are using hardware, chances are it is diverting client
traffic
> > by IP
> > > > address.  If the clients that browse the pages are from sizeable
> > businesses
> > > > and they have proxy servers to the Internet, then you may see
"pooling"
> > to
> > > > one server instead of an even distribution across all servers.  BTW,
> > cable
> > > > modem providers use proxy servers as well.
> > > >
> > > > Check the web logs of all servers to see if  there is an equal
> > distribution.
> > > >
> > > > I hope this helps.
> > >
> >
> --------------------------------------------------------------------------
> > --
> > > > ---------------------------------------
> > > >
> > > > C:\>tracert www.maxim-ic.com
> > > >
> > > > Tracing route to www.maxim-ic.com [209.1.238.246]
> > > > over a maximum of 30 hops:
> > > >
> > > >   1    16 ms    15 ms    16 ms  24.2.41.1
> > > >   2   <10 ms    16 ms   <10 ms  10.0.191.1
> > > >   3   <10 ms    16 ms    16 ms  10.0.188.1
> > > >   4   <10 ms    15 ms   <10 ms  c1-pos5-0.nrflva1.home.net
[24.7.73.109]
> > > >   5   <10 ms   <10 ms    16 ms  c1-pos3-1.washdc1.home.net
[24.7.68.118]
> > > >   6    16 ms    15 ms    16 ms  bb1-se4-1-0.exds-dc.nap.home.net
> > > > [24.7.73.94]
> > > >   7    16 ms    15 ms    16 ms  ibr01-f11-0-0.hrnd01.exodus.net
> > > > [209.185.249.105
> > > > ]
> > > >   8    15 ms    16 ms    16 ms  dcr04-p0-1.hrnd01.exodus.net
> > > > [216.33.203.149]
> > > >   9    15 ms    16 ms    16 ms  bbr02-g4-0.hrnd01.exodus.net
> > > > [216.33.203.126]
> > > >  10    78 ms    93 ms    79 ms  bbr02-p5-0.sntc02.exodus.net
> > [216.32.173.13]
> > > >  11    78 ms    78 ms    94 ms  bbr01-g3-0.sntc02.exodus.net
> > > > [216.33.154.131]
> > > >  12    78 ms    94 ms    78 ms  bbr01-p5-0.sntc01.exodus.net
> > > > [209.185.249.109]
> > > >  13    94 ms    94 ms    78 ms  dcr03-g6-0.sntc01.exodus.net
> > [216.33.147.1]
> > > >  14    78 ms    94 ms    78 ms  rsm02-vlan919.sntc01.exodus.net
> > > > [216.33.147.170]
> > > >
> > > >  15   172 ms    94 ms    93 ms  209.1.238.246
> > > >
> > > > Trace complete.
> > > >
> > >
> >
> --------------------------------------------------------------------------
> > --
> > > > -------------------------------------
> > > >
> > > > C:\>nslookup maxim-ic.com
> > > > Server:  proxy2.nwptn1.va.home.com
> > > > Address:  24.2.48.34
> > > >
> > > > Non-authoritative answer:
> > > > Name:    maxim-ic.com
> > > > Address:  209.1.238.246
> > > >
> > > >
> > > > C:\>nslookup microsoft.com
> > > > Server:  proxy2.nwptn1.va.home.com
> > > > Address:  24.2.48.34
> > > >
> > > > Non-authoritative answer:
> > > > Name:    microsoft.com
> > > > Addresses:  207.46.130.14, 207.46.130.149, 207.46.130.45,
207.46.131.137
> > > >           207.46.131.30
> > > >
> > > >
> > > >
> > >
> >
>

HWG: hwg-servers mailing list archives, maintained by Webmasters @ IWA