home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.misc
- Path: sparky!uunet!cs.utexas.edu!torn!nott!bnrgate!bcars6a8!bcara347!marmen
- From: marmen@bcara347.bnr.ca (Rob Marmen 1532773)
- Subject: Re: NIS - Problems on what systems?
- Message-ID: <1993Jan25.145138.5289@bcars6a8.bnr.ca>
- Summary: One Nasty Prob
- Sender: usenet@bcars6a8.bnr.ca (Use Net)
- Nntp-Posting-Host: bcara347
- Organization: Bell-Northern Research, Ottawa, Canada
- References: <947@wavefront.wti.com>
- Distribution: usa
- Date: Mon, 25 Jan 1993 14:51:38 GMT
- Lines: 87
-
- In article <947@wavefront.wti.com>, kate@wavefront.wti.com (Kathy Samec) writes:
- > What sorts of problems have any of you seen with a heterogeneous network
- > and Different YP(NIS) servers? Is it really all it's hyped up to be?
- >
-
- We have a mix of HPs, Suns and IBMs, and have several NIS domains. The
- largest domain services about 4000 machines and has about 80 servers.
-
- The biggest problem that I have is the inefficiency of the getpw* and
- getgr* routines by all the vendors. We setup accounts as individual
- '+' entries in the passwd file. Consequently, a getpwuid call will
- generate a NIS request for each entry in the passwd file. This causes
- slowdowns for the user. In some cases, telnets can timeout before a
- user complete the login process.
-
- The vendors position is that the software is working as designed, and
- they don't see any reason to change.
-
- Here is the output of a network trace showing the usage on a
- particular subnet. The traffic was logged around 10 AM. The duration
- was three minutes.
-
- 4804 NFS
- 2345 YP
- 1351 XWIN
- 1169 UDP
- 538 RSTAT
- 345 TCP
- 161 RLOGIN
- 122 ARP
- 81 DLC,
- 40 ATP
- 35 RPC
- 30 RIP
- 19 SNMP
- 14 PMAP
- 12 MOUNT
- 9 RTMP
- 6 Telnet
- 4 Ethertype=8005
- 4 DSAP=FC,
- 3 Ethertype=8019
- 2 ECHO
- 1 BOOTP
- ------------------
- 11,095 Total
-
- Please note that about 21 percent of all lan traffic is just NIS/YP
- traffic.
-
- If we further breakdown the NIS traffic by type, then we get the
- following distribution.
-
- 1172 OK (replies from the yp slave)
- 912 bnrottawa.passwd.byname (direct key lookups)
- 234 bnrottawa.services.byname (getservbyname call ) see note
- 11 bnrottawa.netgroup.byhost (direct key lookups)
- 10 bnrottawa.hosts.byname (direct key lookups)
- 4 bnrottawa.hosts.byaddr (direct key lookups)
- 1 service (domain request)
- 1 nothing (do nothing)
- -------------------------------
- 2345 Total
-
- note: Suns do sequential searches of the services.byname map. This is
- replaced by a single lookup on HPs. The spike in
- services.byname traffic is due to the Sun workstations.
-
- From the table, 78 percent of all NIS requests were for a single key
- lookup from the passwd map. If 21 percent of all packets were NIS
- related, then passwd specific lookups make up for around 16 percent
- of the total number of packets in the snapshot.
-
- This is surprising since we always suspected that the services.byname
- traffic was responsible for the majority of the lan traffic.
-
- NIS would be more efficient if the routines were rewritten to minimize
- network traffic.
-
- Hope that this helps.
-
- Rob...
-
- -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
- | Robert Marmen marmen@bnr.ca OR |
- | Bell Northern Research marmen%bnr.ca@cunyvm.cuny.edu |
- | (613) 763-8244 My opinions are my own, not BNRs |
-