| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 4098.1 | DNS is slower than local MIR | CUJO::HILL | Dan Hill-Net.Mgt.-Customer Resident | Wed Dec 02 1992 14:46 | 14 |
| Hi, Dave,
My experience has indicated that DNS V1.1 is roughly 5 times slower
than a local MIR. This will, of course, depend on how fast your DNS
servers are, but you can still use it as a rule of thumb.
V2.0 of DNS should be (I'm told) about 20% faster than V1.1. I have yet
to test it.
Regarding the domain errors, could you be more specific? What kind of
errors? Can you provide an example of your hierarchy and which domains
are giving you problems?
-Dan
|
| 4098.2 | dns server=mcc??? | CTHQ::WOODCOCK | | Wed Dec 02 1992 14:54 | 11 |
| > My experience has indicated that DNS V1.1 is roughly 5 times slower
> than a local MIR. This will, of course, depend on how fast your DNS
> servers are, but you can still use it as a rule of thumb.
Dan, in this comparison was the DNS server resident on the MCC node (eg. a
private namespace) or did the lookups have to go elsewhere? The reason I ask
is that I have a private ns and if the local mir works better I may contemplate
a change.
thanks,
brad...
|
| 4098.3 | What exactly is "5 times" | TOOK::LYONS | Ahh, but fortunately, I have the key to escape reality. | Wed Dec 02 1992 15:24 | 10 |
| Making the statement that Local MIR is 5 time faster is not really
helpfull unless you also add some description about how that number
was arrived at. Local MIR may be faster then DNS whne there are only a
few objects, but what about for MIR's the have 1000 objects, 10000
objects, or event 100,000 objects. Local MIR performance drops off
quickly when the size gets into the 1000 object size, and performance
changes is quite noticable. Also, are the numbers the same (in
magnatude, at least, for object creates, updates, reads, and deletes?
Dave
|
| 4098.4 | True on ultrix, but not on VMS | TOOK::GUERTIN | Don't waffle (unless you want to) | Wed Dec 02 1992 16:09 | 13 |
| Actually Dave, I've only seen the performance of the Local MIR drop off
the way you've described on Ultrix. On VMS, since we use ISAM file
structures, the performance curve is very flat. This is mainly because
they are tuned to run as large files, so small files read relatively
slower than they might ordinarily, and larger files read faster than
they might ordinarily, resulting in a very flat curve. Write and
Delete operations (i.e., Register, Deregister) on VMS Local MIRs will
still slow down at a steeper slope for larger files, but those
operations are done much less frequently than reads (Directory
requests). On Ultrix, the drop off in performance is more dramatic.
-Matt.
|
| 4098.5 | Cases | TOOK::MINTZ | Erik Mintz | Wed Dec 02 1992 16:13 | 17 |
| Consider several cases:
1) Local namespace (no DECdns involved)
2) DNS server on the same system as MCC
3) DNS server on the same LAN
4) Remote DNS server
2 may well be slower than either 1 or 3. Both
DECdns servers and DECmcc are fairly resource intensive,
so it may not be best to force them to compete for resources
on the same system.
The speed of 3 and 4 depend on the network latency, the speed of the
machine hosting the DNS server, and the number of requests from
other sources placing a load on the server.
|
| 4098.6 | DNS namespace on remote servers | CUJO::HILL | Dan Hill-Net.Mgt.-Customer Resident | Tue Dec 08 1992 23:15 | 17 |
| Brad,
To answer your question, the DNS server nodes are uVAXes, neither one
being a DECmcc node. I suspect that a 40+ VUP system with a second
generation Ethernet controller might shave this factor of 5 down
considerably, but I'd be willing to bet that the local RMS-based MIR
will still be noticably faster.
I have almost 1500 entities in my local MIR and it screams past what
DNS provided. Matt has done some testing with Ultrix, as I'll bet Dave
has, also, hence their comments about large files and slow performance.
RMS is a different story.
Additionally, Erik is right on the money when it comes to performance
degradation. I've tried it each way. #1 is best, #4 is worst.
-Dan
|