Capacity & RAM

Re: Capacity & RAM

From:

Date:

Time:

Hi all,

I wanted to post an update to this thread with my observations of our particular experiment:

>From: "Peter Lalor" <plalor@infoasis.com>
>Date: Mon, 8 Nov 1999 11:52:51 -0800
>
> >From: "Men & Mice Support" <cbuxton@menandmice.com>
> >
> >At 10:16 AM -0600 11/5/99, Jerry Pasker-Systems Admin. wrote:
> > >I've been running
> > >my primary on a 6220 CD, (75Mhz 603, and the slowest PowerMac ever made)
> > >with 24MB of RAM (12 allocated to QDNS Pro 2.2.1) and have found that under
> > >a pretty good load (3-10 quirries per seccond) less RAM is better than more
> > >RAM. I'm only assuming that it's faster to search 1.5MB of cached records,
> > >and then resolve a name using an external server(s) than it is to sort
> > >through 11.5MB of cached records.
> > >
> > >I'm sure there has to be a guideline to setting RAM on different speeds of
> > >machines. Or is more allways better? Can anyone clarify this?
> >
> >In theory, more RAM is better, up to the point where the available
> >RAM is not all used. Beyond that, more RAM is simply unused. You can
> >check the free memory reading in the status window.
> >
> >However, that is a slow machine.
> >
> >Keep in mind that, no matter how fast the machine, there will be a
> >point where more RAM will slow it down. This is true of any type of
> >system that stores data in a table in RAM. Furthermore, when
> >resolving outside queries, QuickDNS Pro can also find the same
> >answers by the usual resolution routine, so the point of inefficiency
> >can be even lower, especially for slow machines.
>
>This is a very interesting point: Where does more processor outweigh
>more cache? It would seem that the answer depends on many factors,
>including the number of clients hitting the box (more will improve
>cache efficiencies) and how the root servers are doing.
>
>We had a larger cache on our secondary machine than our primary, but
>it's cache hit ratio was only about 30%. Our primary saw about a 75%
>cache hit ratio, with a smaller cache of about 8 Mb. Now, we're
>experimenting with a 55 Mb cache on the primary and the cache hit
>ratio is 80% after less than 24 hours in operation--but it shows 49
>Mb free RAM (the machine has too much RAM, so what the hell). That
>server hosts well over 200 zones.
>
>But is it faster? Who knows? We'll have to watch it as it builds it's
>cache. DNS is not something I'm particularly inclined to assign much
>processor to, but it absolutely needs to be "fast enough".
>
>Peter Lalor Infoasis

We've been experimenting with a 6150 (66 MHz PPC 601, 256k L2 cache) with 72 Mb RAM, OS 9. The machine is dedicated to QuickDNS Pro, hosts a couple hundred domains, and is the primary client DNS server.

The machine has been running for one week, and in that week has sent roughly 900,000 responses, 73% from cache. Free memory is 27 Mb of the 55 Mb allocated to the QuickDNS Pro app. At peak times, the machine is sustaining about 220 kb/s of DNS traffic.

Seat-of-the-pants feeling: the hardware is fast enough. We might swap it for something like a 7200/120 after further observation, but it seems pretty damn snappy.

The only problem we've had is with Funnel Web analysis of very large Web server logs. We find that Funnel Web generates a DNS load exceeding 1 mb/s and that it effectively brings QuickDNS to it's knees, no matter what hardware we've run it on. That said, large Unix iron doesn't appreciate it much either.

HTH.

Peter Lalor Infoasis
plalor@infoasis.com The San Francisco Bay Area's Macintosh
415-459-7991 Consultant and Internet Service Provider
415-459-7992 fax http://www.infoasis.com/

Sales:
888/4-DPOINT

Support:
858/452-3696