home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.isis
- Path: sparky!uunet!mcsun!dxcern!sun2!fabien
- From: fabien@sun2.cern.ch (Fabien Collin)
- Subject: ISIS 3.0.5 & Round trip time (Bug ?)
- Message-ID: <1992Jul21.135255.28380@dxcern.cern.ch>
- Sender: fabien@sun2 (Fabien Collin)
- Organization: European Laboratory for Particule Physics (CERN) Switzerland
- Distribution: comp
- Date: Tue, 21 Jul 1992 13:52:55 GMT
- Lines: 60
-
- I did a benchmark to evaluate ISIS round trip time.
- I used ISIS 3.0.5 with two SUN 4C under SunOS 4.1.1 connected by
- Ethernet.
-
- I have two processes (a master and a slave) which belong to the same
- group (Bypass facility is used).
- The following operation is repeated n times :
-
- The master send a message (size = s) with a cbcast to the slave.
- The master then waits for a reply. The slave replies by sending
- a message (size = r).
-
- I did a test with s = r and the mean time for the round trip was
- measured, I obtained the following results :
-
-
- Msg Size (s,r) (bytes) Round trip time (ms)
- 1 6
- 14624 55
- 14625 550
-
- 29424 600
- 29425 1100
-
- 44224 1150
- 44225 1650
-
- 59024 1750
- 59025 2250
-
- ...
-
- Between 1 and 14624, the time varies linearly from 6 ms to 55 ms.
- Suddenly, between 14624 and 14625, it takes 0.5 seconds more
- to do the round trip. The same phenomenon is repeated each time
- you add 14800 bytes.
-
- I suppose the discontinuity is due to the size of a buffer.
- But half a second each 14800 bytes seems to be enormous to me.
- Is it a bug ?
- If it is, is it fixed in 3.0.6 ?
- It it isn't, how do you explain this huge amount of time ?
-
- PS: I obtained the same results with a abcast. It is not surprising
- since I used only 1 destination and only this event exists in the system.
-
- PS2: Same results with two asynchronous cbcast (or abcast).
-
-
- Cheers,
-
- Fabien
-
- --
- -------------------------------------------------------------------------
- COLLIN Fabien | Computing & Networking Division
- Email: fabien@sun2.cern.ch | Building 31, 3-013
- Voice: + 41-22-767-2487 | CERN 1211 Geneva 23
- Fax : + 41-22-767-7155 | Switzerland
- -------------------------------------------------------------------------
-