THANK you, Christof!
I'm going to have to lock my office door and unplug the phone so I can
work through your incredibly informative post...and hope to understand
it all.
I, very much, appreciate your advice and input!
Mike
-------- Original Message --------
Subject: Re: [NF] Intel i5 vs. Xeon CPU for a data server
From: Christof Wollenhaupt <[email protected]>
To: [email protected]
Date: 6/19/2013 3:33 AM
CPU usage isn't a valid measurement to compare desktop and server CPUs. The
same application might use 5% on a Core i5, but only 2% on a Xeon E7 due to
triple the cache, better conflict management, better predictive queues,
etc. A Xeon would still be twice as fast in this case. A core isn't a
separate CPU. A number of parts of the CPU do not exist once per core and
are shared among cores. Most importantly, the cache and the bus are shared.
Before you do any guessing, you could rent a server in a data center for a
month and do some load testing yourself. Amazon rents Xeon machines by the
hour for a few cents per hour. In a data center you get a i5 for less than
$100 a month and Xeon for less than $200.
With a database server I'd pay probably pay more attention to storage,
memory and connectivity.
Instead of local storage, I'd probably get a SAN connected via multipath GB
network. There're inexpensive SANs available from Qnap, Synology, and the
like. Most of them offer replication between two instances. If the database
server performs replication, that's extra load and bandwidth taken from
read-only, in memory queries that the server could do instead.
If your database application has lots of writes you have to pay attention
to IOPS on the storage system. 200 clients on a single mirrored disk system
with 150 IOPS means that a write operation will take about a second.
Replace it with a RAID that has many spindles, you increase IOPS linearly.
IOPS are the reason that most server disks are still small (600 GB, 1.2
TB), because in a server you want more disks, nor larger disks.
Closely related is the topic of alignment. In a database server the
database block size MUST align with the cluster block size and location of
file systems and disk. Otherwise you might end up with a system that has to
read/write multiple clusters for every database block. That's especially
important as some file systems use significantly larger cluster sizes of up
to 2 MB or with virtualized drives.
When you test network performance keep in mind that bandwidth is only part
of the equations. The other important measure is packets per second (PPS).
Every switch/router/hub has a limit of how many IP packets it can manage
per second. Higher PPS values are often the reason that a business switch
is so much more expensive than a consumer device even if both have GB ports.
--
Christof
--- StripMime Report -- processed MIME parts ---
multipart/alternative
text/plain (text body -- kept)
text/html
---
[excessive quoting removed by server]
_______________________________________________
Post Messages to: [email protected]
Subscription Maintenance: http://mail.leafe.com/mailman/listinfo/profox
OT-free version of this list: http://mail.leafe.com/mailman/listinfo/profoxtech
Searchable Archive: http://leafe.com/archives/search/profox
This message:
http://leafe.com/archives/byMID/profox/[email protected]
** All postings, unless explicitly stated otherwise, are the opinions of the
author, and do not constitute legal or medical advice. This statement is added
to the messages for those lawyers who are too stupid to see the obvious.
